<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"
	xmlns:content="http://purl.org/rss/1.0/modules/content/"
	xmlns:wfw="http://wellformedweb.org/CommentAPI/"
	xmlns:dc="http://purl.org/dc/elements/1.1/"
	xmlns:atom="http://www.w3.org/2005/Atom"
	xmlns:sy="http://purl.org/rss/1.0/modules/syndication/"
	xmlns:slash="http://purl.org/rss/1.0/modules/slash/"
	>

<channel>
	<title>Case Studies - Chimes AI</title>
	<atom:link href="https://chimes.ai/category/case-studies/feed/" rel="self" type="application/rss+xml" />
	<link>https://chimes.ai</link>
	<description>用 AI 實踐 ESG 企業永續</description>
	<lastBuildDate>Mon, 22 Dec 2025 10:15:36 +0000</lastBuildDate>
	<language>zh-TW</language>
	<sy:updatePeriod>
	hourly	</sy:updatePeriod>
	<sy:updateFrequency>
	1	</sy:updateFrequency>
	<generator>https://wordpress.org/?v=6.2.8</generator>

<image>
	<url>https://chimes.ai/wp-content/uploads/2021/05/cropped-tukeylogo-32x32.png</url>
	<title>Case Studies - Chimes AI</title>
	<link>https://chimes.ai</link>
	<width>32</width>
	<height>32</height>
</image> 
	<item>
		<title>Chimes AI 參與「韌性臺灣異地強固資通訊計畫」，推動政府數位轉型與風險預警應用</title>
		<link>https://chimes.ai/2025/04/29/chimes-ai-resilient-taiwan-digital-transformation-risk-prediction/?utm_source=rss&#038;utm_medium=rss&#038;utm_campaign=chimes-ai-resilient-taiwan-digital-transformation-risk-prediction</link>
					<comments>https://chimes.ai/2025/04/29/chimes-ai-resilient-taiwan-digital-transformation-risk-prediction/#respond</comments>
		
		<dc:creator><![CDATA[Chimes AI]]></dc:creator>
		<pubDate>Tue, 29 Apr 2025 10:00:36 +0000</pubDate>
				<category><![CDATA[AI Innovations]]></category>
		<category><![CDATA[Case Studies]]></category>
		<category><![CDATA[Industry Insights]]></category>
		<category><![CDATA[Sustainability]]></category>
		<category><![CDATA[AI in Public Policy]]></category>
		<category><![CDATA[Disaster Risk Prediction]]></category>
		<category><![CDATA[Government Digital Transformation]]></category>
		<category><![CDATA[Smart Governance]]></category>
		<category><![CDATA[Sustainable Development]]></category>
		<category><![CDATA[Taiwan Resilience Program]]></category>
		<guid isPermaLink="false">https://chimes.ai/?p=6671</guid>

					<description><![CDATA[<p>Chimes AI 今日受邀參加由內政部消防署舉辦的「韌性臺</p>
<div><a href="https://chimes.ai/2025/04/29/chimes-ai-resilient-taiwan-digital-transformation-risk-prediction/" class="exp-read-more exp-read-more-underlined">Read More</a></div>
<p>The post <a href="https://chimes.ai/2025/04/29/chimes-ai-resilient-taiwan-digital-transformation-risk-prediction/">Chimes AI 參與「韌性臺灣異地強固資通訊計畫」，推動政府數位轉型與風險預警應用</a> first appeared on <a href="https://chimes.ai">Chimes AI</a>.</p>]]></description>
										<content:encoded><![CDATA[<p><span style="font-weight: 400;">Chimes AI 今日受邀參加由內政部消防署舉辦的「韌性臺灣異地強固資通訊計畫」資訊新知研習會。該活動聚焦於提升臺灣在災害防救，及社會安全方面的數位應用能力。Chimes AI 商務開發總監簡仁瑋分享了政府數位轉型的資料思維，並說明如何透過大數據與生成式 AI 技術進行災害預防與風險管理。展示了多項創新應用，其中包括：</span></p>
<ul>
<li style="font-weight: 400;" aria-level="1"><b>OHCA（到院前心肺休止）風險地圖</b><span style="font-weight: 400;">與</span><b>住宅火災風險地圖</b><span style="font-weight: 400;">：這些地圖透過風險指數高低預測，幫助強化高風險地區的緊急救護策略，並透過優化政府機關系統整合，提高緊急醫療救護品質。</span>&nbsp;</li>
<li style="font-weight: 400;" aria-level="1"><b>兒少保護風險預測</b><span style="font-weight: 400;">：目前全國每年超過 20,000 件兒少家暴案件，平均一位社工需處理 60 案，其中 30% 屬於家暴回頭客。Chimes AI 運用大數據技術建立家暴風險預測模型，能將新通報案件分類為高、中、低再受暴指數，幫助社會及家庭署有效分配資源與服務強度，強化臺灣社會安全網絡建置。</span>&nbsp;</li>
<li style="font-weight: 400;" aria-level="1"><b>颱洪致災停電風險預警</b><span style="font-weight: 400;">：整合颱風預測資料、歷史停電資料等數據建立預測模型，並結合生成式 AI 技術，打造「颱洪致災停電小幫手」。該工具能快速生成颱風侵台的停電預測，識別高風險區域，並提供台電公司警控中心防災建議，助力提升電網韌性與供電品質。</span>&nbsp;</li>
<li style="font-weight: 400;" aria-level="1"><b>長期 2.0 照顧計畫</b><span style="font-weight: 400;">：透過長照服務歷程資料庫進行服務流程資料探勘，理解個案因失能程度、家庭照顧者與經濟狀況的差異需求，該模型能推薦最適合的家庭長照服務組合，提供切合需求的個案與家屬照護方案，顯著提升長照服務的效能與精準度。</span></li>
</ul>
<p><span style="font-weight: 400;">Chimes AI 致力於用數據與 AI 技術提升災害預警與社會安全防護的能力。通過跨部門的技術整合與創新應用，為臺灣的韌性建設與社會福祉做出更大貢獻。</span></p>
<p><span style="font-weight: 400;">如需了解更多數據分析應用案例，請造訪公司官網：</span><a href="https://chimes.ai"><span style="font-weight: 400;">https://chimes.ai</span></a></p><p>The post <a href="https://chimes.ai/2025/04/29/chimes-ai-resilient-taiwan-digital-transformation-risk-prediction/">Chimes AI 參與「韌性臺灣異地強固資通訊計畫」，推動政府數位轉型與風險預警應用</a> first appeared on <a href="https://chimes.ai">Chimes AI</a>.</p>]]></content:encoded>
					
					<wfw:commentRss>https://chimes.ai/2025/04/29/chimes-ai-resilient-taiwan-digital-transformation-risk-prediction/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
			</item>
		<item>
		<title>Chimes AI 於《ESG遠見》分享 AI 結合 ESG 的雙重效益，助力實現製造業與公共服務的永續願景</title>
		<link>https://chimes.ai/2025/02/27/chimes-ai-powering-sustainability-with-ai-and-esg/?utm_source=rss&#038;utm_medium=rss&#038;utm_campaign=chimes-ai-powering-sustainability-with-ai-and-esg</link>
					<comments>https://chimes.ai/2025/02/27/chimes-ai-powering-sustainability-with-ai-and-esg/#respond</comments>
		
		<dc:creator><![CDATA[Chimes AI]]></dc:creator>
		<pubDate>Thu, 27 Feb 2025 07:10:49 +0000</pubDate>
				<category><![CDATA[AI Innovations]]></category>
		<category><![CDATA[Case Studies]]></category>
		<category><![CDATA[Industry Insights]]></category>
		<category><![CDATA[Manufacturing Technology]]></category>
		<category><![CDATA[Sustainability]]></category>
		<category><![CDATA[Thought Leadership]]></category>
		<category><![CDATA[AI for ESG]]></category>
		<category><![CDATA[公共服務 AI]]></category>
		<category><![CDATA[智慧製造]]></category>
		<category><![CDATA[永續發展]]></category>
		<category><![CDATA[遠見 ESG 共好圈]]></category>
		<guid isPermaLink="false">https://chimes.ai/?p=6757</guid>

					<description><![CDATA[<p>詠鋐智能（Chimes AI, Inc.）創辦人暨執行長謝宗</p>
<div><a href="https://chimes.ai/2025/02/27/chimes-ai-powering-sustainability-with-ai-and-esg/" class="exp-read-more exp-read-more-underlined">Read More</a></div>
<p>The post <a href="https://chimes.ai/2025/02/27/chimes-ai-powering-sustainability-with-ai-and-esg/">Chimes AI 於《ESG遠見》分享 AI 結合 ESG 的雙重效益，助力實現製造業與公共服務的永續願景</a> first appeared on <a href="https://chimes.ai">Chimes AI</a>.</p>]]></description>
										<content:encoded><![CDATA[<p><span style="font-weight: 400;">詠鋐智能（Chimes AI, Inc.）創辦人暨執行長謝宗震博士受邀參與「遠見 ESG 共好圈」線上知識分享會，與業界共同探討 AI 與 ESG 的深度結合如何改變永續發展格局，助力企業與政府部門實現商業與社會價值的雙贏。</span></p>
<p><b>AI 與 ESG 的完美結合：從預警到解題</b></p>
<p><span style="font-weight: 400;">謝宗震博士在分享中表示：「AI 與 ESG 的結合已是現在進行式，從設備異常預測到智慧維護管理，AI 正以前所未有的速度推動永續轉型。」他歸納了 AI 在 ESG 中的四大應用場景，包括設備異常預測、品質提升與能源優化、智慧維護管理以及知識圖譜與安全治理。</span></p>
<p><b>實務應用：台塑與鋼鐵業的成功案例</b></p>
<p><span style="font-weight: 400;">在台塑石化的應用中， Chimes AI 透過即時感測與 AI 演算法模型，針對冷卻水汞、壓縮機與風車等設備建立智慧監診機制，提前辨識設備潛在問題並預防故障發生。這項創新技術不僅有效節省了超過 20% 的維護成本，更顯著降低了化學品外洩風險，進一步提升環境保護與工業安全水準，為企業實現雙重效益樹立新標杆。</span></p>
<p><span style="font-weight: 400;">而在鋼鐵業的應用中， Chimes AI 運用 AI 技術開發最佳化升溫策略，確保產品品質的同時顯著減少能源浪費，每年節省數百萬度電並大幅降低碳排放。謝宗震博士指出：「設備的折舊隨著使用次數而增加，但 AI 模型卻隨著更多運轉數據而更精準。」透過 AI 優化，企業得以持續找出降低能耗的最佳模式，進一步實現節能減碳目標。</span></p>
<p><b>社會安全網的進化：AI 在公共服務中的應用</b></p>
<p><span style="font-weight: 400;">除了產業應用， Chimes AI 亦在兒少家暴與長照服務中發揮關鍵作用。透過 AI 建立風險預警模型，社會工作者得以即時識別高風險個案，提前介入，減少再度受虐的機率。此外，AI 對長照服務進行優化，縮短照護磨合時間，提升服務效率與品質。</span></p>
<p><span style="font-weight: 400;">在消防安全領域，詠鋐智能協助高雄市建置火災風險地圖，整合火警資料與建築特性，優化宣導策略，將高風險區域火災發生率降低 50%。</span></p>
<p><b>AI 驅動永續未來</b></p>
<p><span style="font-weight: 400;">AI 是企業與政府預警 ESG 風險的核心工具，通過分析數據，AI 能幫助企業提前發現問題，制定明確的 ESG 藍圖，從而實現超前部署，補足商業與社會價值的空隙。 Chimes AI 將持續推動 AI 在永續發展中的應用，為全球的永續未來注入動能。</span></p>
<p><span style="font-weight: 400;">如需了解更多 Tukey 產品導入方案或邀約線上產品演示，請造訪公司官網：</span><a href="http://www.chimes.ai/"><span style="font-weight: 400;">www.chimes.ai</span></a></p><p>The post <a href="https://chimes.ai/2025/02/27/chimes-ai-powering-sustainability-with-ai-and-esg/">Chimes AI 於《ESG遠見》分享 AI 結合 ESG 的雙重效益，助力實現製造業與公共服務的永續願景</a> first appeared on <a href="https://chimes.ai">Chimes AI</a>.</p>]]></content:encoded>
					
					<wfw:commentRss>https://chimes.ai/2025/02/27/chimes-ai-powering-sustainability-with-ai-and-esg/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
			</item>
		<item>
		<title>Chimes AI 助力中鋼碳素實現節能減碳突破，亞灣計畫成果亮相「Meet亞灣新創大南方」</title>
		<link>https://chimes.ai/2024/08/24/chimes-ai-csc-energy-savings-carbon-reduction-meet-south-taiwan/?utm_source=rss&#038;utm_medium=rss&#038;utm_campaign=chimes-ai-csc-energy-savings-carbon-reduction-meet-south-taiwan</link>
					<comments>https://chimes.ai/2024/08/24/chimes-ai-csc-energy-savings-carbon-reduction-meet-south-taiwan/#respond</comments>
		
		<dc:creator><![CDATA[Chimes AI]]></dc:creator>
		<pubDate>Sat, 24 Aug 2024 03:54:25 +0000</pubDate>
				<category><![CDATA[Case Studies]]></category>
		<category><![CDATA[Industry Insights]]></category>
		<category><![CDATA[News]]></category>
		<category><![CDATA[Sustainability]]></category>
		<category><![CDATA[NoCodeAI]]></category>
		<category><![CDATA[SmartManufacturing]]></category>
		<category><![CDATA[亞灣2.0]]></category>
		<category><![CDATA[加熱爐]]></category>
		<category><![CDATA[節能減碳]]></category>
		<guid isPermaLink="false">https://chimes.ai/?p=6313</guid>

					<description><![CDATA[<p>Chimes AI (詠鋐智能) 今日於南臺灣規模最大的創新</p>
<div><a href="https://chimes.ai/2024/08/24/chimes-ai-csc-energy-savings-carbon-reduction-meet-south-taiwan/" class="exp-read-more exp-read-more-underlined">Read More</a></div>
<p>The post <a href="https://chimes.ai/2024/08/24/chimes-ai-csc-energy-savings-carbon-reduction-meet-south-taiwan/">Chimes AI 助力中鋼碳素實現節能減碳突破，亞灣計畫成果亮相「Meet亞灣新創大南方」</a> first appeared on <a href="https://chimes.ai">Chimes AI</a>.</p>]]></description>
										<content:encoded><![CDATA[<p><span class="s1">Chimes AI (詠鋐智能) 今日於南臺灣規模最大的創新活動「Meet亞灣新創大南方」展會上，隆重展示其在「高雄亞灣創新生態系計畫」中的最新技術成果。透過尖端的人工智慧技術，Chimes AI 成功協助中鋼碳素在製程節能減碳方面取得了重大突破，電加熱爐製程節電達32%，相當於每年節省40萬度電，減少約200噸的二氧化碳排放。此技術方案具有高度的可複製性，未來可推廣至全台鋼鐵及石化廠，為這些高耗能產業帶來顯著的能源優化與減碳效益。</span></p>
<p class="p1"><span class="s1">Chimes AI 的資深資料科學家洪國彭在展會中表示：「透過我們的 AI 模組，中鋼碳素不僅在製程中實現了顯著的節能效果，還大幅降低了整體碳排放量。這項技術將為全台鋼鐵與石化廠提供重要的節能減碳新方向。」此次技術應用中，Chimes AI 的能源AI模組運用了先進的 AutoML 和 PSO 粒子群演算法，成功優化操作參數，使能源消耗降至最低，同時確保產品品質合規。</span></p>
<p class="p1"><span class="s1">Chimes AI 致力於降低 AI 應用的建置門檻，透過無程式碼（No-Code）平台，使得非技術背景的工程師也能輕鬆運用 AI 工具，實現製程的智慧化轉型。中鋼碳素此次成功應用 AI 技術，不僅為該公司在能源管理和碳排放控制領域樹立了新標桿，也為全台相關產業的升級轉型提供了可行方案。</span></p><p>The post <a href="https://chimes.ai/2024/08/24/chimes-ai-csc-energy-savings-carbon-reduction-meet-south-taiwan/">Chimes AI 助力中鋼碳素實現節能減碳突破，亞灣計畫成果亮相「Meet亞灣新創大南方」</a> first appeared on <a href="https://chimes.ai">Chimes AI</a>.</p>]]></content:encoded>
					
					<wfw:commentRss>https://chimes.ai/2024/08/24/chimes-ai-csc-energy-savings-carbon-reduction-meet-south-taiwan/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
			</item>
		<item>
		<title>台塑導入 No-Code AI 工具 Tukey！智慧保養模型讓製程更穩定</title>
		<link>https://chimes.ai/2021/11/20/%e5%8f%b0%e5%a1%91%e5%b0%8e%e5%85%a5-no-code-ai-%e5%b7%a5%e5%85%b7-tukey%ef%bc%81%e6%99%ba%e6%85%a7%e4%bf%9d%e9%a4%8a%e6%a8%a1%e5%9e%8b%e8%ae%93%e8%a3%bd%e7%a8%8b%e6%9b%b4%e7%a9%a9%e5%ae%9a/?utm_source=rss&#038;utm_medium=rss&#038;utm_campaign=%25e5%258f%25b0%25e5%25a1%2591%25e5%25b0%258e%25e5%2585%25a5-no-code-ai-%25e5%25b7%25a5%25e5%2585%25b7-tukey%25ef%25bc%2581%25e6%2599%25ba%25e6%2585%25a7%25e4%25bf%259d%25e9%25a4%258a%25e6%25a8%25a1%25e5%259e%258b%25e8%25ae%2593%25e8%25a3%25bd%25e7%25a8%258b%25e6%259b%25b4%25e7%25a9%25a9%25e5%25ae%259a</link>
					<comments>https://chimes.ai/2021/11/20/%e5%8f%b0%e5%a1%91%e5%b0%8e%e5%85%a5-no-code-ai-%e5%b7%a5%e5%85%b7-tukey%ef%bc%81%e6%99%ba%e6%85%a7%e4%bf%9d%e9%a4%8a%e6%a8%a1%e5%9e%8b%e8%ae%93%e8%a3%bd%e7%a8%8b%e6%9b%b4%e7%a9%a9%e5%ae%9a/#respond</comments>
		
		<dc:creator><![CDATA[Chimes AI]]></dc:creator>
		<pubDate>Sat, 20 Nov 2021 08:55:38 +0000</pubDate>
				<category><![CDATA[Case Studies]]></category>
		<category><![CDATA[Manufacturing Technology]]></category>
		<category><![CDATA[AIinManufacturing]]></category>
		<category><![CDATA[OperationalExcellence]]></category>
		<category><![CDATA[Predictive Maintenance]]></category>
		<category><![CDATA[SmartManufacturing]]></category>
		<guid isPermaLink="false">https://chimes.ai/?p=6129</guid>

					<description><![CDATA[<p>台灣石化業龍頭台塑企業長年來積極推展數位轉型，近期投入人工智</p>
<div><a href="https://chimes.ai/2021/11/20/%e5%8f%b0%e5%a1%91%e5%b0%8e%e5%85%a5-no-code-ai-%e5%b7%a5%e5%85%b7-tukey%ef%bc%81%e6%99%ba%e6%85%a7%e4%bf%9d%e9%a4%8a%e6%a8%a1%e5%9e%8b%e8%ae%93%e8%a3%bd%e7%a8%8b%e6%9b%b4%e7%a9%a9%e5%ae%9a/" class="exp-read-more exp-read-more-underlined">Read More</a></div>
<p>The post <a href="https://chimes.ai/2021/11/20/%e5%8f%b0%e5%a1%91%e5%b0%8e%e5%85%a5-no-code-ai-%e5%b7%a5%e5%85%b7-tukey%ef%bc%81%e6%99%ba%e6%85%a7%e4%bf%9d%e9%a4%8a%e6%a8%a1%e5%9e%8b%e8%ae%93%e8%a3%bd%e7%a8%8b%e6%9b%b4%e7%a9%a9%e5%ae%9a/">台塑導入 No-Code AI 工具 Tukey！智慧保養模型讓製程更穩定</a> first appeared on <a href="https://chimes.ai">Chimes AI</a>.</p>]]></description>
										<content:encoded><![CDATA[<p id="8980" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">台灣石化業龍頭台塑企業長年來積極推展數位轉型，近期投入人工智慧（ AI ）的技術開發與運用。在「產銷優化、品質確保、智慧保養、工安環保、降低成本」五大面向，持續深化應用，提升效益。其中，台塑公司設備保養管理負責部門，是負責製程設備完整性及可靠性，以確保製程產線穩定運轉，藉由AI技術運用提升設備可靠度，降低無預期設備故障。</p>
<p id="12ac" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">無預警停機造成的影響不僅是生產損失，亦有可能衍生工安環保問題，因此石化製程對於設備完整性要求較其他產業為高，因此傳統預知保養已難在符合現況製程穩定性要求。</p>
<p id="b50e" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">傳統上保養策略為定期保養、預防保養、預知檢測等保養策略，隨者科技發展製程設備相對複雜且專業，且每位保養人員負責設備機種與廠牌繁多，傳統保養策略已無法符合保養作業需求，台塑近年來逐步導入設備智慧監診系統，以感應器收取機台運轉資訊，並結合專家知識、修繕履歷建立 AI 模型，已實踐設備極早期預警功能，在設備進入劣化狀態之際，提前發出預警，以降低非預期停機維修的損失。</p>
<h1 id="b3a1" class="nq nr fr be ns nt nu nv nw nx ny nz oa ob oc od oe of og oh oi oj ok ol om on bj" data-selectable-paragraph="">AI 模型的維運難題</h1>
<p id="c375" class="pw-post-body-paragraph mo mp fr mq b mr oo mt mu mv op mx my mz oq nb nc nd or nf ng nh os nj nk nl fk bj" data-selectable-paragraph="">台塑公司的設備智慧監診系統，主要由資訊部門協助開發，設備管理部門使用。在多年運行後發現，發現AI 模型在長期使用之後，預測效果不如預期發生資料飄移（Data drift）現象，也就是在設備狀態改變，可能是因設備逢大週期整修零件大幅汰換後，其設備表徵(震動、溫度、壓力)呈現改變，往往會跟先前的資料分布不一樣。此時正在線上使用的 AI 模型泛用性隨之變差。</p>
<p id="be26" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">每當模型表現衰退時，會需要資訊人員協助重新訓練模型（Model retrain）。面對製程大量設備，每台機器就需要建立一個模型，AI模型適當調校即為一大問題；再者，每台設備的情況各自不同，若有特殊停機或是維修的事件，在建立模型時需要特別額外做資料處理，如果不是對機台的狀況非常熟悉的人，會需要大量的跨部門溝通協作。</p>
<h1 id="857b" class="nq nr fr be ns nt nu nv nw nx ny nz oa ob oc od oe of og oh oi oj ok ol om on bj" data-selectable-paragraph="">Tukey 降低 AI 模型建置門檻，大幅縮短模型開發的時間</h1>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu aob"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/0*NEWm9sveZBkcQFB1.png 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/0*NEWm9sveZBkcQFB1.png 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/0*NEWm9sveZBkcQFB1.png 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/0*NEWm9sveZBkcQFB1.png 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/0*NEWm9sveZBkcQFB1.png 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/0*NEWm9sveZBkcQFB1.png 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/0*NEWm9sveZBkcQFB1.png 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/0*NEWm9sveZBkcQFB1.png 640w, https://miro.medium.com/v2/resize:fit:720/0*NEWm9sveZBkcQFB1.png 720w, https://miro.medium.com/v2/resize:fit:750/0*NEWm9sveZBkcQFB1.png 750w, https://miro.medium.com/v2/resize:fit:786/0*NEWm9sveZBkcQFB1.png 786w, https://miro.medium.com/v2/resize:fit:828/0*NEWm9sveZBkcQFB1.png 828w, https://miro.medium.com/v2/resize:fit:1100/0*NEWm9sveZBkcQFB1.png 1100w, https://miro.medium.com/v2/resize:fit:1400/0*NEWm9sveZBkcQFB1.png 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/0*NEWm9sveZBkcQFB1.png" alt="" width="700" height="338" /></picture></div>
</div><figcaption class="mi mj mk lt lu ml mm be b bf z dw" data-selectable-paragraph="">(圖說)台塑保養中心智慧監診系統，系統流程圖。</figcaption></figure>
<p id="42c0" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">2021年6月，台塑導入由 Chimes AI 詠鋐智能所開發的 No-Code 平台 Tukey，讓最熟悉機台狀況，但沒有程式開發能力的設備保養工程師自行建置模型。</p>
<p id="ca46" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">在過往，當模型的表現衰退時，需要由保養部門委託資訊部門重新訓練模型，由雙方所建立的工作小組共同協作，建立一個模型需要花費六個月的時間。</p>
<p id="a9e7" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">在導入 Tukey 之後，模型再訓練的工作由各保養廠的維修專員自行執行，在短短三個月之內，已經建置並上線 400 多個模型，大幅縮短模型開發的時間。</p>
<h1 id="1151" class="nq nr fr be ns nt nu nv nw nx ny nz oa ob oc od oe of og oh oi oj ok ol om on bj" data-selectable-paragraph="">Tukey 提供跨組織的溝通橋樑，提升 AI 模型準確率</h1>
<p id="da9e" class="pw-post-body-paragraph mo mp fr mq b mr oo mt mu mv op mx my mz oq nb nc nd or nf ng nh os nj nk nl fk bj" data-selectable-paragraph="">台塑主管提到，由於設備工程師是負責設備維運人員，最熟悉自己負責的機台狀況。相較於過往，由資訊人員協助所建立的模型，在資料清洗的步驟上，更能考量設備穩定操作各項因素各項設備特徵呈現，更能建立符合設備診斷AI模型。體現資料科學的名言：資料決定模型 表現上限，演算法逼近上限。</p>
<p id="01c6" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">過往在資料清理步驟，多是設備人員跟由資訊人員溝通後，以 Python 撰寫程式碼，做客製化的清理。Tukey 可紀錄下每個資料清理的步驟，協助跨部門溝通時，有共同的對話基礎。在導入 Tukey 的三個月內，提升 AI 模型準確率達 5%，有效提升生產線的穩定度。</p>
<p id="c902" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">台塑集團AI 佈局逐漸成形，台塑導入 No-Code AI 工具，拉近 AI 與第一線設備工程師的距離，提升工作效率。未來，台塑也期望將這一套工作流程拓展到海外工廠，擴展 AI 佈局，降低維護成本並提升工廠工作安全。</p>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu aob"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/0*E1jExumSJleQhGOj.jpg 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/0*E1jExumSJleQhGOj.jpg 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/0*E1jExumSJleQhGOj.jpg 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/0*E1jExumSJleQhGOj.jpg 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/0*E1jExumSJleQhGOj.jpg 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/0*E1jExumSJleQhGOj.jpg 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/0*E1jExumSJleQhGOj.jpg 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/0*E1jExumSJleQhGOj.jpg 640w, https://miro.medium.com/v2/resize:fit:720/0*E1jExumSJleQhGOj.jpg 720w, https://miro.medium.com/v2/resize:fit:750/0*E1jExumSJleQhGOj.jpg 750w, https://miro.medium.com/v2/resize:fit:786/0*E1jExumSJleQhGOj.jpg 786w, https://miro.medium.com/v2/resize:fit:828/0*E1jExumSJleQhGOj.jpg 828w, https://miro.medium.com/v2/resize:fit:1100/0*E1jExumSJleQhGOj.jpg 1100w, https://miro.medium.com/v2/resize:fit:1400/0*E1jExumSJleQhGOj.jpg 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/0*E1jExumSJleQhGOj.jpg" alt="" width="700" height="394" /></picture></div>
</div>
</figure><p>The post <a href="https://chimes.ai/2021/11/20/%e5%8f%b0%e5%a1%91%e5%b0%8e%e5%85%a5-no-code-ai-%e5%b7%a5%e5%85%b7-tukey%ef%bc%81%e6%99%ba%e6%85%a7%e4%bf%9d%e9%a4%8a%e6%a8%a1%e5%9e%8b%e8%ae%93%e8%a3%bd%e7%a8%8b%e6%9b%b4%e7%a9%a9%e5%ae%9a/">台塑導入 No-Code AI 工具 Tukey！智慧保養模型讓製程更穩定</a> first appeared on <a href="https://chimes.ai">Chimes AI</a>.</p>]]></content:encoded>
					
					<wfw:commentRss>https://chimes.ai/2021/11/20/%e5%8f%b0%e5%a1%91%e5%b0%8e%e5%85%a5-no-code-ai-%e5%b7%a5%e5%85%b7-tukey%ef%bc%81%e6%99%ba%e6%85%a7%e4%bf%9d%e9%a4%8a%e6%a8%a1%e5%9e%8b%e8%ae%93%e8%a3%bd%e7%a8%8b%e6%9b%b4%e7%a9%a9%e5%ae%9a/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
			</item>
		<item>
		<title>用 Tukey 執行精準行銷，以潛在客戶名單推薦為例</title>
		<link>https://chimes.ai/2021/11/01/%e7%94%a8-tukey-%e5%9f%b7%e8%a1%8c%e7%b2%be%e6%ba%96%e8%a1%8c%e9%8a%b7-%e4%bb%a5%e6%bd%9b%e5%9c%a8%e5%ae%a2%e6%88%b6%e5%90%8d%e5%96%ae%e6%8e%a8%e8%96%a6%e7%82%ba%e4%be%8b/?utm_source=rss&#038;utm_medium=rss&#038;utm_campaign=%25e7%2594%25a8-tukey-%25e5%259f%25b7%25e8%25a1%258c%25e7%25b2%25be%25e6%25ba%2596%25e8%25a1%258c%25e9%258a%25b7-%25e4%25bb%25a5%25e6%25bd%259b%25e5%259c%25a8%25e5%25ae%25a2%25e6%2588%25b6%25e5%2590%258d%25e5%2596%25ae%25e6%258e%25a8%25e8%2596%25a6%25e7%2582%25ba%25e4%25be%258b</link>
					<comments>https://chimes.ai/2021/11/01/%e7%94%a8-tukey-%e5%9f%b7%e8%a1%8c%e7%b2%be%e6%ba%96%e8%a1%8c%e9%8a%b7-%e4%bb%a5%e6%bd%9b%e5%9c%a8%e5%ae%a2%e6%88%b6%e5%90%8d%e5%96%ae%e6%8e%a8%e8%96%a6%e7%82%ba%e4%be%8b/#respond</comments>
		
		<dc:creator><![CDATA[Chimes AI]]></dc:creator>
		<pubDate>Mon, 01 Nov 2021 08:50:43 +0000</pubDate>
				<category><![CDATA[Case Studies]]></category>
		<category><![CDATA[Enterprise AI]]></category>
		<category><![CDATA[Martech]]></category>
		<category><![CDATA[No-Code AI Tools]]></category>
		<category><![CDATA[Predictive Modeling]]></category>
		<guid isPermaLink="false">https://chimes.ai/?p=6124</guid>

					<description><![CDATA[<p>原文連結：掌握資料科學的四大分析步驟，啟動 AI 專案 數位</p>
<div><a href="https://chimes.ai/2021/11/01/%e7%94%a8-tukey-%e5%9f%b7%e8%a1%8c%e7%b2%be%e6%ba%96%e8%a1%8c%e9%8a%b7-%e4%bb%a5%e6%bd%9b%e5%9c%a8%e5%ae%a2%e6%88%b6%e5%90%8d%e5%96%ae%e6%8e%a8%e8%96%a6%e7%82%ba%e4%be%8b/" class="exp-read-more exp-read-more-underlined">Read More</a></div>
<p>The post <a href="https://chimes.ai/2021/11/01/%e7%94%a8-tukey-%e5%9f%b7%e8%a1%8c%e7%b2%be%e6%ba%96%e8%a1%8c%e9%8a%b7-%e4%bb%a5%e6%bd%9b%e5%9c%a8%e5%ae%a2%e6%88%b6%e5%90%8d%e5%96%ae%e6%8e%a8%e8%96%a6%e7%82%ba%e4%be%8b/">用 Tukey 執行精準行銷，以潛在客戶名單推薦為例</a> first appeared on <a href="https://chimes.ai">Chimes AI</a>.</p>]]></description>
										<content:encoded><![CDATA[<div class="ab ca">
<div class="ch bg ew ex ey ez">
<p id="6728" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">原文連結：<a class="af mn" href="https://www.bnext.com.tw/article/65653/chimes-ai-tukey" target="_blank" rel="noopener ugc nofollow">掌握資料科學的四大分析步驟，啟動 AI 專案</a></p>
<p id="20d6" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">數位科技的運用，不僅驅動智慧化生活，企業也加快推動數位轉型，人工智慧（AI）的應用更是愈趨廣泛。舉凡語音助理、手機美拍、人臉辨識、個人會行銷，都是日常生活中常見的應用。這些新穎的技術，不只讓生活更方便，也悄悄地影響了多數職場工作者的工作內容。對非電腦科學背景的職場工作者來說，能不能善用人工智慧、與人工智慧協同合作，將是影響未來職場競爭力的關鍵。</p>
<p id="f446" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">不會寫程式，也能做 AI 專案嗎？全球知名資訊科技研究顧問機構 Gartner 近日發佈的一份<a class="af mn" href="https://www.gartner.com/en/newsroom/press-releases/2021-06-10-gartner-says-the-majority-of-technology-products-and-services-will-be-built-by-professionals-outside-of-it-by-2024" target="_blank" rel="noopener ugc nofollow">文章</a>指出，在 2024 年將有 80% 的科技產品與服務出自於非 IT 技術專業人士，這項變革背後最大的推手即是無程式碼工具（No-Code AI tool）。只要瞭解基礎的資料分析流程與原理，運用這類的 AI 建模軟體，即可快速建立 AI 模型。</p>
<h1 id="5127" class="nq nr fr be ns nt nu nv nw nx ny nz oa ob oc od oe of og oh oi oj ok ol om on bj" data-selectable-paragraph="">資料科學專案的分析步驟</h1>
<p id="8f6b" class="pw-post-body-paragraph mo mp fr mq b mr oo mt mu mv op mx my mz oq nb nc nd or nf ng nh os nj nk nl fk bj" data-selectable-paragraph="">以精準行銷的 AI 專案為例：某雜誌擁有大量的訂戶，平時也有經營社群，並且推廣課程。今年推出了一套針對中高階主管的全新的商管課程，希望能夠透過精準投放，提高課程的購買率。然而行銷的總預算有限，在茫茫的會員人群中，要如何篩選出成交機率高的會員，提升效率與利潤，是行銷部門的首要之務。</p>
<p id="602c" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">要達到上述目的，需要打造「潛在客戶名單推薦系統」，作法如下：首先，要將訂戶資料與社群會員資料做整併，進行初步的資料探索。在這個步驟，我們可能發現有少部分客戶的年齡被誤植成負數（資料探索）。為了不影響後續建立模型後的準確率，我們將這幾筆資料予以刪除（資料清理）。接著以 AI 建模軟體建立購課成交率的預測模型（建立模型），再從中選出表現最佳的模型，介接至潛在客戶名單推薦系統，對所有客戶名單進行購課成交率預測，針對成交率高的客戶，進行簡訊或電話行銷（模型部署、實際應用）。</p>
<p id="7e23" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">從以上的範例可以發現，資料科學的工作流程大致可以歸納成以下四大步驟：</p>
<h2 id="382f" class="amb nr fr be ns aid amc aie nw aig amd aih oa mz ame yd yg nd amf yh yk nh amg yl yo amh bj" data-selectable-paragraph="">一、資料探索</h2>
<p id="2e05" class="pw-post-body-paragraph mo mp fr mq b mr oo mt mu mv op mx my mz oq nb nc nd or nf ng nh os nj nk nl fk bj" data-selectable-paragraph="">對手上的分析繪製簡單的統計圖表，並且觀察個欄位的統計量。其目的在於了解資料的分佈是否與認知相符，查看離群值的分佈，找出關聯性，從圖表找出洞察，建立並測試假說。<br />
在這個步驟，經常查看的統計量有：平均值（Mean）、中位數（Median）、眾數（Mode）、最小值（Min）、最大值（Max）、範圍（Range）、四分位差（Quartiles）、變異數（Variance）、標準差（Standard deviation）。<br />
圖表方面，常見的單變量的圖表有直方圖（Histogram）、柱狀圖（Bar chart）；雙變量的圖表則是會看散佈圖（Scatter plot）、箱型圖（Box plot）、熱力圖（Heat map）。</p>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu aob"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/0*QDhXPA4lWSvYOYYO.jpg 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/0*QDhXPA4lWSvYOYYO.jpg 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/0*QDhXPA4lWSvYOYYO.jpg 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/0*QDhXPA4lWSvYOYYO.jpg 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/0*QDhXPA4lWSvYOYYO.jpg 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/0*QDhXPA4lWSvYOYYO.jpg 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/0*QDhXPA4lWSvYOYYO.jpg 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/0*QDhXPA4lWSvYOYYO.jpg 640w, https://miro.medium.com/v2/resize:fit:720/0*QDhXPA4lWSvYOYYO.jpg 720w, https://miro.medium.com/v2/resize:fit:750/0*QDhXPA4lWSvYOYYO.jpg 750w, https://miro.medium.com/v2/resize:fit:786/0*QDhXPA4lWSvYOYYO.jpg 786w, https://miro.medium.com/v2/resize:fit:828/0*QDhXPA4lWSvYOYYO.jpg 828w, https://miro.medium.com/v2/resize:fit:1100/0*QDhXPA4lWSvYOYYO.jpg 1100w, https://miro.medium.com/v2/resize:fit:1400/0*QDhXPA4lWSvYOYYO.jpg 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/0*QDhXPA4lWSvYOYYO.jpg" alt="" width="700" height="553" /></picture></div>
</div><figcaption class="mi mj mk lt lu ml mm be b bf z dw" data-selectable-paragraph="">圖說：直方圖可以查看單一欄位的分佈。以此圖為例，訂戶的訂閱期數（subscribe_time）在 50 期以下佔多數。</figcaption></figure>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu aob"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/0*c6nGRQ49ghcFhBE8.jpg 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/0*c6nGRQ49ghcFhBE8.jpg 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/0*c6nGRQ49ghcFhBE8.jpg 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/0*c6nGRQ49ghcFhBE8.jpg 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/0*c6nGRQ49ghcFhBE8.jpg 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/0*c6nGRQ49ghcFhBE8.jpg 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/0*c6nGRQ49ghcFhBE8.jpg 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/0*c6nGRQ49ghcFhBE8.jpg 640w, https://miro.medium.com/v2/resize:fit:720/0*c6nGRQ49ghcFhBE8.jpg 720w, https://miro.medium.com/v2/resize:fit:750/0*c6nGRQ49ghcFhBE8.jpg 750w, https://miro.medium.com/v2/resize:fit:786/0*c6nGRQ49ghcFhBE8.jpg 786w, https://miro.medium.com/v2/resize:fit:828/0*c6nGRQ49ghcFhBE8.jpg 828w, https://miro.medium.com/v2/resize:fit:1100/0*c6nGRQ49ghcFhBE8.jpg 1100w, https://miro.medium.com/v2/resize:fit:1400/0*c6nGRQ49ghcFhBE8.jpg 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/0*c6nGRQ49ghcFhBE8.jpg" alt="" width="700" height="553" /></picture></div>
</div><figcaption class="mi mj mk lt lu ml mm be b bf z dw" data-selectable-paragraph="">圖說：訂戶資料長條圖顯示續訂戶約250筆，是新訂戶約60筆的4倍。</figcaption></figure>
<h2 id="b18e" class="amb nr fr be ns aid amc aie nw aig amd aih oa mz ame yd yg nd amf yh yk nh amg yl yo amh bj" data-selectable-paragraph="">二、資料清理</h2>
<p id="5398" class="pw-post-body-paragraph mo mp fr mq b mr oo mt mu mv op mx my mz oq nb nc nd or nf ng nh os nj nk nl fk bj" data-selectable-paragraph="">在資料探索階段，可能會發現許多內容不合理的數據，或是遺失值。此時需要仔細檢查為什麼會收到這樣的資料，並針對不同的情境，做出相對應的資料清理動作。<br />
像是發現某些訂戶資料的年齡為負數時，深入去檢查原始資料庫的資料，發現是雜誌社的工作人員登打的時候，將客戶的出生日期打錯了，因此後續才會有有不合理的年齡資訊。此時可以考慮將該筆資料移除，維持資料的正確性。</p>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu aob"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/0*HJ5HaSUCfhho1eKH.jpg 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/0*HJ5HaSUCfhho1eKH.jpg 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/0*HJ5HaSUCfhho1eKH.jpg 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/0*HJ5HaSUCfhho1eKH.jpg 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/0*HJ5HaSUCfhho1eKH.jpg 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/0*HJ5HaSUCfhho1eKH.jpg 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/0*HJ5HaSUCfhho1eKH.jpg 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/0*HJ5HaSUCfhho1eKH.jpg 640w, https://miro.medium.com/v2/resize:fit:720/0*HJ5HaSUCfhho1eKH.jpg 720w, https://miro.medium.com/v2/resize:fit:750/0*HJ5HaSUCfhho1eKH.jpg 750w, https://miro.medium.com/v2/resize:fit:786/0*HJ5HaSUCfhho1eKH.jpg 786w, https://miro.medium.com/v2/resize:fit:828/0*HJ5HaSUCfhho1eKH.jpg 828w, https://miro.medium.com/v2/resize:fit:1100/0*HJ5HaSUCfhho1eKH.jpg 1100w, https://miro.medium.com/v2/resize:fit:1400/0*HJ5HaSUCfhho1eKH.jpg 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/0*HJ5HaSUCfhho1eKH.jpg" alt="" width="700" height="553" /></picture></div>
</div><figcaption class="mi mj mk lt lu ml mm be b bf z dw" data-selectable-paragraph="">圖說：從直方圖發現訂戶年齡（age）為負值</figcaption></figure>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu aob"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/0*0eASwxNDL4yauDnm.jpg 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/0*0eASwxNDL4yauDnm.jpg 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/0*0eASwxNDL4yauDnm.jpg 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/0*0eASwxNDL4yauDnm.jpg 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/0*0eASwxNDL4yauDnm.jpg 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/0*0eASwxNDL4yauDnm.jpg 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/0*0eASwxNDL4yauDnm.jpg 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/0*0eASwxNDL4yauDnm.jpg 640w, https://miro.medium.com/v2/resize:fit:720/0*0eASwxNDL4yauDnm.jpg 720w, https://miro.medium.com/v2/resize:fit:750/0*0eASwxNDL4yauDnm.jpg 750w, https://miro.medium.com/v2/resize:fit:786/0*0eASwxNDL4yauDnm.jpg 786w, https://miro.medium.com/v2/resize:fit:828/0*0eASwxNDL4yauDnm.jpg 828w, https://miro.medium.com/v2/resize:fit:1100/0*0eASwxNDL4yauDnm.jpg 1100w, https://miro.medium.com/v2/resize:fit:1400/0*0eASwxNDL4yauDnm.jpg 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/0*0eASwxNDL4yauDnm.jpg" alt="" width="700" height="553" /></picture></div>
</div><figcaption class="mi mj mk lt lu ml mm be b bf z dw" data-selectable-paragraph="">圖說：將年齡為負值的資料刪除。</figcaption></figure>
<h2 id="e9b1" class="amb nr fr be ns aid amc aie nw aig amd aih oa mz ame yd yg nd amf yh yk nh amg yl yo amh bj" data-selectable-paragraph="">三、建立模型</h2>
<p id="d620" class="pw-post-body-paragraph mo mp fr mq b mr oo mt mu mv op mx my mz oq nb nc nd or nf ng nh os nj nk nl fk bj" data-selectable-paragraph="">資料清理完畢之後，只要訓練資料集與演算法之後，即可建立 AI 模型。目前市面上的 AI 建模軟體已經內建 Auto ML 技術，使用者毋須一一調整參數，軟體會自動進行參數最佳化，將最佳的結果回傳。</p>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu aob"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/0*T-JFs18HKv5nZJ7O.jpg 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/0*T-JFs18HKv5nZJ7O.jpg 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/0*T-JFs18HKv5nZJ7O.jpg 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/0*T-JFs18HKv5nZJ7O.jpg 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/0*T-JFs18HKv5nZJ7O.jpg 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/0*T-JFs18HKv5nZJ7O.jpg 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/0*T-JFs18HKv5nZJ7O.jpg 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/0*T-JFs18HKv5nZJ7O.jpg 640w, https://miro.medium.com/v2/resize:fit:720/0*T-JFs18HKv5nZJ7O.jpg 720w, https://miro.medium.com/v2/resize:fit:750/0*T-JFs18HKv5nZJ7O.jpg 750w, https://miro.medium.com/v2/resize:fit:786/0*T-JFs18HKv5nZJ7O.jpg 786w, https://miro.medium.com/v2/resize:fit:828/0*T-JFs18HKv5nZJ7O.jpg 828w, https://miro.medium.com/v2/resize:fit:1100/0*T-JFs18HKv5nZJ7O.jpg 1100w, https://miro.medium.com/v2/resize:fit:1400/0*T-JFs18HKv5nZJ7O.jpg 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/0*T-JFs18HKv5nZJ7O.jpg" alt="" width="700" height="449" /></picture></div>
</div><figcaption class="mi mj mk lt lu ml mm be b bf z dw" data-selectable-paragraph="">圖說：將模型的基礎資訊設定完畢，即可自動建立模型。</figcaption></figure>
<h2 id="a21c" class="amb nr fr be ns aid amc aie nw aig amd aih oa mz ame yd yg nd amf yh yk nh amg yl yo amh bj" data-selectable-paragraph="">四、模型部署、實際應用</h2>
<p id="a9f5" class="pw-post-body-paragraph mo mp fr mq b mr oo mt mu mv op mx my mz oq nb nc nd or nf ng nh os nj nk nl fk bj" data-selectable-paragraph="">上述的步驟可以多次反覆執行，建立多個 AI 模型。從中挑選出表現較佳的模型，將之部署上線，開始實際運用到工作場域中。</p>
<p id="2265" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">上述的範例中，行銷人員每月可將新收集到會員資料匯入「潛在客戶名單推薦系統」，系統即回傳成交機率較高的客戶名單。行銷人員即可用這份名單執行後續的促銷活動。</p>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu aob"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/0*30oeHkcXJwFDSIhH.jpg 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/0*30oeHkcXJwFDSIhH.jpg 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/0*30oeHkcXJwFDSIhH.jpg 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/0*30oeHkcXJwFDSIhH.jpg 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/0*30oeHkcXJwFDSIhH.jpg 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/0*30oeHkcXJwFDSIhH.jpg 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/0*30oeHkcXJwFDSIhH.jpg 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/0*30oeHkcXJwFDSIhH.jpg 640w, https://miro.medium.com/v2/resize:fit:720/0*30oeHkcXJwFDSIhH.jpg 720w, https://miro.medium.com/v2/resize:fit:750/0*30oeHkcXJwFDSIhH.jpg 750w, https://miro.medium.com/v2/resize:fit:786/0*30oeHkcXJwFDSIhH.jpg 786w, https://miro.medium.com/v2/resize:fit:828/0*30oeHkcXJwFDSIhH.jpg 828w, https://miro.medium.com/v2/resize:fit:1100/0*30oeHkcXJwFDSIhH.jpg 1100w, https://miro.medium.com/v2/resize:fit:1400/0*30oeHkcXJwFDSIhH.jpg 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/0*30oeHkcXJwFDSIhH.jpg" alt="" width="700" height="534" /></picture></div>
</div><figcaption class="mi mj mk lt lu ml mm be b bf z dw" data-selectable-paragraph="">圖說：將 Tukey 模型導入「潛在客戶名單推薦系統」，系統即回傳成交機率較高的客戶名單，提供行銷人員做後續使用。</figcaption></figure>
<h1 id="d331" class="nq nr fr be ns nt nu nv nw nx ny nz oa ob oc od oe of og oh oi oj ok ol om on bj" data-selectable-paragraph="">AI 時代的行銷利器：Tukey</h1>
<p id="5be5" class="pw-post-body-paragraph mo mp fr mq b mr oo mt mu mv op mx my mz oq nb nc nd or nf ng nh os nj nk nl fk bj" data-selectable-paragraph="">對於行銷、業務人員而言，不外乎想要讓更多人可以認識自家產品，提高潛在客戶的成交意願，進而提升訂單成交率。由 Chimes AI 詠鋐智能所研發的企業級 AI 建模與管理平台 Tukey ，可提供精準投放、商品推薦、挖掘潛在商機、商品銷量預測等各種行銷面向上的決策輔助。</p>
<p id="d3e7" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">Tukey 的 No-Code AI 產品設計，提供非機器學習演算法專家 (譬如：銀行理財專員、電商營運專員) 簡單直覺的操作介面，讓直接面對營運問題的一線人員，迅速完成 AI 模型建置，提升工作效率。Tukey 也可以完整追溯資料專案的資料歷程，在需要跨部門團隊協作的場合，彌平認知落差，亦可與他人進行跨平台的協同運作，增加工作的一致性與正確性，進而提升工作效率。</p>
<p id="7291" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">Tukey 目前已經被台塑公司採用，並且被納入台灣人工智慧學校的高階經理人班教材。想了解更多，歡迎報名《數位時代》推出的「<a class="af mn" href="https://edm.managertoday.com.tw/2021data-science/" target="_blank" rel="noopener ugc nofollow">資料科學概念系列課</a>」！立即打造你的第一個 AI 行銷專案吧！</p>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu aob"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/0*c506yS8VTEvhpC4P.png 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/0*c506yS8VTEvhpC4P.png 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/0*c506yS8VTEvhpC4P.png 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/0*c506yS8VTEvhpC4P.png 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/0*c506yS8VTEvhpC4P.png 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/0*c506yS8VTEvhpC4P.png 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/0*c506yS8VTEvhpC4P.png 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/0*c506yS8VTEvhpC4P.png 640w, https://miro.medium.com/v2/resize:fit:720/0*c506yS8VTEvhpC4P.png 720w, https://miro.medium.com/v2/resize:fit:750/0*c506yS8VTEvhpC4P.png 750w, https://miro.medium.com/v2/resize:fit:786/0*c506yS8VTEvhpC4P.png 786w, https://miro.medium.com/v2/resize:fit:828/0*c506yS8VTEvhpC4P.png 828w, https://miro.medium.com/v2/resize:fit:1100/0*c506yS8VTEvhpC4P.png 1100w, https://miro.medium.com/v2/resize:fit:1400/0*c506yS8VTEvhpC4P.png 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/0*c506yS8VTEvhpC4P.png" alt="" width="700" height="394" /></picture></div>
</div>
</figure>
</div>
</div><p>The post <a href="https://chimes.ai/2021/11/01/%e7%94%a8-tukey-%e5%9f%b7%e8%a1%8c%e7%b2%be%e6%ba%96%e8%a1%8c%e9%8a%b7-%e4%bb%a5%e6%bd%9b%e5%9c%a8%e5%ae%a2%e6%88%b6%e5%90%8d%e5%96%ae%e6%8e%a8%e8%96%a6%e7%82%ba%e4%be%8b/">用 Tukey 執行精準行銷，以潛在客戶名單推薦為例</a> first appeared on <a href="https://chimes.ai">Chimes AI</a>.</p>]]></content:encoded>
					
					<wfw:commentRss>https://chimes.ai/2021/11/01/%e7%94%a8-tukey-%e5%9f%b7%e8%a1%8c%e7%b2%be%e6%ba%96%e8%a1%8c%e9%8a%b7-%e4%bb%a5%e6%bd%9b%e5%9c%a8%e5%ae%a2%e6%88%b6%e5%90%8d%e5%96%ae%e6%8e%a8%e8%96%a6%e7%82%ba%e4%be%8b/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
			</item>
		<item>
		<title>用 Tukey 打造設備性能即時監控系統，以風力發電機為例</title>
		<link>https://chimes.ai/2021/09/07/%e7%94%a8-tukey-%e6%89%93%e9%80%a0%e8%a8%ad%e5%82%99%e6%80%a7%e8%83%bd%e5%8d%b3%e6%99%82%e7%9b%a3%e6%8e%a7%e7%b3%bb%e7%b5%b1-%e4%bb%a5%e9%a2%a8%e5%8a%9b%e7%99%bc%e9%9b%bb%e6%a9%9f%e7%82%ba%e4%be%8b/?utm_source=rss&#038;utm_medium=rss&#038;utm_campaign=%25e7%2594%25a8-tukey-%25e6%2589%2593%25e9%2580%25a0%25e8%25a8%25ad%25e5%2582%2599%25e6%2580%25a7%25e8%2583%25bd%25e5%258d%25b3%25e6%2599%2582%25e7%259b%25a3%25e6%258e%25a7%25e7%25b3%25bb%25e7%25b5%25b1-%25e4%25bb%25a5%25e9%25a2%25a8%25e5%258a%259b%25e7%2599%25bc%25e9%259b%25bb%25e6%25a9%259f%25e7%2582%25ba%25e4%25be%258b</link>
					<comments>https://chimes.ai/2021/09/07/%e7%94%a8-tukey-%e6%89%93%e9%80%a0%e8%a8%ad%e5%82%99%e6%80%a7%e8%83%bd%e5%8d%b3%e6%99%82%e7%9b%a3%e6%8e%a7%e7%b3%bb%e7%b5%b1-%e4%bb%a5%e9%a2%a8%e5%8a%9b%e7%99%bc%e9%9b%bb%e6%a9%9f%e7%82%ba%e4%be%8b/#respond</comments>
		
		<dc:creator><![CDATA[Chimes AI]]></dc:creator>
		<pubDate>Tue, 07 Sep 2021 08:41:31 +0000</pubDate>
				<category><![CDATA[Case Studies]]></category>
		<category><![CDATA[Predictive Maintenance]]></category>
		<category><![CDATA[Renewable Energy]]></category>
		<guid isPermaLink="false">https://chimes.ai/?p=6120</guid>

					<description><![CDATA[<p>台灣的風能資源豐富舉世公認，擁有陸域風場 30 處，陸域風機</p>
<div><a href="https://chimes.ai/2021/09/07/%e7%94%a8-tukey-%e6%89%93%e9%80%a0%e8%a8%ad%e5%82%99%e6%80%a7%e8%83%bd%e5%8d%b3%e6%99%82%e7%9b%a3%e6%8e%a7%e7%b3%bb%e7%b5%b1-%e4%bb%a5%e9%a2%a8%e5%8a%9b%e7%99%bc%e9%9b%bb%e6%a9%9f%e7%82%ba%e4%be%8b/" class="exp-read-more exp-read-more-underlined">Read More</a></div>
<p>The post <a href="https://chimes.ai/2021/09/07/%e7%94%a8-tukey-%e6%89%93%e9%80%a0%e8%a8%ad%e5%82%99%e6%80%a7%e8%83%bd%e5%8d%b3%e6%99%82%e7%9b%a3%e6%8e%a7%e7%b3%bb%e7%b5%b1-%e4%bb%a5%e9%a2%a8%e5%8a%9b%e7%99%bc%e9%9b%bb%e6%a9%9f%e7%82%ba%e4%be%8b/">用 Tukey 打造設備性能即時監控系統，以風力發電機為例</a> first appeared on <a href="https://chimes.ai">Chimes AI</a>.</p>]]></description>
										<content:encoded><![CDATA[<div class="ab ca">
<div class="ch bg ew ex ey ez">
<p id="8375" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">台灣的風能資源豐富舉世公認，擁有陸域風場 30 處，陸域風機 350 餘台。尤其西部沿海與澎湖地區由於地形效應，冬季東北季風與夏季西南季風特別旺盛，提供發展風力發電之有利條件，據統計 2020 年發電量達 24.33 億度 (2,433GWh)。然而，台灣高溫潮濕的環境，亦使得風機設備在保養維護上面臨挑戰。因此，為了讓風機處於隨時都能夠發電的最佳狀態，需要在風機運轉發生異常的初期，即時通報維修人員進行處理，一方面降低設備損耗，同時也能延長風機的使用壽命，維持風機運轉穩定。</p>
<p id="b521" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">Tukey 為智慧工廠打造的 AI 建模與管理平台，讓設備保養人員在不用寫程式的情況下，輕鬆快速的建立風機設備性能監控 AI 模型。</p>
<h1 id="1479" class="nq nr fr be ns nt nu nv nw nx ny nz oa ob oc od oe of og oh oi oj ok ol om on bj" data-selectable-paragraph="">通過這篇文章，你可以學到</h1>
<ul class="">
<li id="e67f" class="mo mp fr mq b mr oo mt mu mv op mx my mz oq nb nc nd or nf ng nh os nj nk nl ala alb alc bj" data-selectable-paragraph="">透過擬真案例，迅速理解設備性能監測的 End-to-End 建置流程，包含：</li>
<li id="64fe" class="mo mp fr mq b mr ald mt mu mv ale mx my mz alf nb nc nd alg nf ng nh alh nj nk nl ala alb alc bj" data-selectable-paragraph="">執行資料探索分析與資料清洗</li>
<li id="ceaa" class="mo mp fr mq b mr ald mt mu mv ale mx my mz alf nb nc nd alg nf ng nh alh nj nk nl ala alb alc bj" data-selectable-paragraph="">以 AutoML 建立 baseline 模型</li>
<li id="6f38" class="mo mp fr mq b mr ald mt mu mv ale mx my mz alf nb nc nd alg nf ng nh alh nj nk nl ala alb alc bj" data-selectable-paragraph="">基於模型評估指標與圖表，進行模型再調校</li>
<li id="00df" class="mo mp fr mq b mr ald mt mu mv ale mx my mz alf nb nc nd alg nf ng nh alh nj nk nl ala alb alc bj" data-selectable-paragraph="">啟動 Tukey Model API，串接風機性能即時監控看板</li>
</ul>
<h1 id="aaf0" class="nq nr fr be ns nt nu nv nw nx ny nz oa ob oc od oe of og oh oi oj ok ol om on bj" data-selectable-paragraph="">背景知識與分析目標</h1>
<blockquote class="nm nn no">
<p id="6b92" class="mo mp np mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">聲明：本範例資料為開放資料模擬再製，與設備實際規格、台灣風場現況不盡相符。</p>
</blockquote>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu ama"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/1*WiRkd3XgFSGZ9NSzeaN8ow.png 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/1*WiRkd3XgFSGZ9NSzeaN8ow.png 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/1*WiRkd3XgFSGZ9NSzeaN8ow.png 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/1*WiRkd3XgFSGZ9NSzeaN8ow.png 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/1*WiRkd3XgFSGZ9NSzeaN8ow.png 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/1*WiRkd3XgFSGZ9NSzeaN8ow.png 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/1*WiRkd3XgFSGZ9NSzeaN8ow.png 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/1*WiRkd3XgFSGZ9NSzeaN8ow.png 640w, https://miro.medium.com/v2/resize:fit:720/1*WiRkd3XgFSGZ9NSzeaN8ow.png 720w, https://miro.medium.com/v2/resize:fit:750/1*WiRkd3XgFSGZ9NSzeaN8ow.png 750w, https://miro.medium.com/v2/resize:fit:786/1*WiRkd3XgFSGZ9NSzeaN8ow.png 786w, https://miro.medium.com/v2/resize:fit:828/1*WiRkd3XgFSGZ9NSzeaN8ow.png 828w, https://miro.medium.com/v2/resize:fit:1100/1*WiRkd3XgFSGZ9NSzeaN8ow.png 1100w, https://miro.medium.com/v2/resize:fit:1400/1*WiRkd3XgFSGZ9NSzeaN8ow.png 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/1*WiRkd3XgFSGZ9NSzeaN8ow.png" alt="" width="700" height="380" /></picture></div>
</div><figcaption class="mi mj mk lt lu ml mm be b bf z dw" data-selectable-paragraph="">風力發電機資料說明</figcaption></figure>
<p id="c5af" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">自風機 SCADA 系統擷取運轉資料，每十分鐘一筆。如上圖所示，資料包含發電功率、風速、風向、軸承溫度、齒輪箱潤滑油溫度、發電機轉速、風機液壓油壓力與風機液壓油溫度等。</p>
<p id="ea06" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">已知發電功率（GRID_POWER）與風速（AMB_WINDSPEED）有強相關性。希望打造一個風機設備性能監控 AI 模型，監控風機的發電功率是否正常。</p>
<h1 id="1535" class="nq nr fr be ns nt nu nv nw nx ny nz oa ob oc od oe of og oh oi oj ok ol om on bj" data-selectable-paragraph="">分析流程</h1>
<h2 id="e2ce" class="amb nr fr be ns aid amc aie nw aig amd aih oa mz ame yd yg nd amf yh yk nh amg yl yo amh bj" data-selectable-paragraph="">1. 匯入風機運轉資料</h2>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu ami"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/1*dgqS-3RBXDBpgT0J8gjp-A.png 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/1*dgqS-3RBXDBpgT0J8gjp-A.png 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/1*dgqS-3RBXDBpgT0J8gjp-A.png 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/1*dgqS-3RBXDBpgT0J8gjp-A.png 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/1*dgqS-3RBXDBpgT0J8gjp-A.png 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/1*dgqS-3RBXDBpgT0J8gjp-A.png 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/1*dgqS-3RBXDBpgT0J8gjp-A.png 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/1*dgqS-3RBXDBpgT0J8gjp-A.png 640w, https://miro.medium.com/v2/resize:fit:720/1*dgqS-3RBXDBpgT0J8gjp-A.png 720w, https://miro.medium.com/v2/resize:fit:750/1*dgqS-3RBXDBpgT0J8gjp-A.png 750w, https://miro.medium.com/v2/resize:fit:786/1*dgqS-3RBXDBpgT0J8gjp-A.png 786w, https://miro.medium.com/v2/resize:fit:828/1*dgqS-3RBXDBpgT0J8gjp-A.png 828w, https://miro.medium.com/v2/resize:fit:1100/1*dgqS-3RBXDBpgT0J8gjp-A.png 1100w, https://miro.medium.com/v2/resize:fit:1400/1*dgqS-3RBXDBpgT0J8gjp-A.png 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/1*dgqS-3RBXDBpgT0J8gjp-A.png" alt="" width="700" height="395" /></picture></div>
</div>
</figure>
<p>&nbsp;</p>
<h2 id="66a7" class="amb nr fr be ns aid amc aie nw aig amd aih oa mz ame yd yg nd amf yh yk nh amg yl yo amh bj" data-selectable-paragraph="">2. 觀察資料的分佈</h2>
<p id="a825" class="pw-post-body-paragraph mo mp fr mq b mr oo mt mu mv op mx my mz oq nb nc nd or nf ng nh os nj nk nl fk bj" data-selectable-paragraph="">以單維度與雙維度的資料視覺化，進行資料探索，觀察資料趨勢與分佈。</p>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu ami"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/1*CCnBrKYnzqylopFvVIgyzQ.png 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/1*CCnBrKYnzqylopFvVIgyzQ.png 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/1*CCnBrKYnzqylopFvVIgyzQ.png 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/1*CCnBrKYnzqylopFvVIgyzQ.png 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/1*CCnBrKYnzqylopFvVIgyzQ.png 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/1*CCnBrKYnzqylopFvVIgyzQ.png 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/1*CCnBrKYnzqylopFvVIgyzQ.png 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/1*CCnBrKYnzqylopFvVIgyzQ.png 640w, https://miro.medium.com/v2/resize:fit:720/1*CCnBrKYnzqylopFvVIgyzQ.png 720w, https://miro.medium.com/v2/resize:fit:750/1*CCnBrKYnzqylopFvVIgyzQ.png 750w, https://miro.medium.com/v2/resize:fit:786/1*CCnBrKYnzqylopFvVIgyzQ.png 786w, https://miro.medium.com/v2/resize:fit:828/1*CCnBrKYnzqylopFvVIgyzQ.png 828w, https://miro.medium.com/v2/resize:fit:1100/1*CCnBrKYnzqylopFvVIgyzQ.png 1100w, https://miro.medium.com/v2/resize:fit:1400/1*CCnBrKYnzqylopFvVIgyzQ.png 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/1*CCnBrKYnzqylopFvVIgyzQ.png" alt="" width="700" height="395" /></picture></div>
</div>
</figure>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu amj"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/1*VJ6nGdhn6jpUZ_P51vwqeA.png 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/1*VJ6nGdhn6jpUZ_P51vwqeA.png 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/1*VJ6nGdhn6jpUZ_P51vwqeA.png 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/1*VJ6nGdhn6jpUZ_P51vwqeA.png 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/1*VJ6nGdhn6jpUZ_P51vwqeA.png 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/1*VJ6nGdhn6jpUZ_P51vwqeA.png 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/1*VJ6nGdhn6jpUZ_P51vwqeA.png 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/1*VJ6nGdhn6jpUZ_P51vwqeA.png 640w, https://miro.medium.com/v2/resize:fit:720/1*VJ6nGdhn6jpUZ_P51vwqeA.png 720w, https://miro.medium.com/v2/resize:fit:750/1*VJ6nGdhn6jpUZ_P51vwqeA.png 750w, https://miro.medium.com/v2/resize:fit:786/1*VJ6nGdhn6jpUZ_P51vwqeA.png 786w, https://miro.medium.com/v2/resize:fit:828/1*VJ6nGdhn6jpUZ_P51vwqeA.png 828w, https://miro.medium.com/v2/resize:fit:1100/1*VJ6nGdhn6jpUZ_P51vwqeA.png 1100w, https://miro.medium.com/v2/resize:fit:1400/1*VJ6nGdhn6jpUZ_P51vwqeA.png 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/1*VJ6nGdhn6jpUZ_P51vwqeA.png" alt="" width="700" height="377" /></picture></div>
</div>
</figure>
<p id="66a7" class="amb nr fr be ns aid amc aie nw aig amd aih oa mz ame yd yg nd amf yh yk nh amg yl yo amh bj">
<h2 id="8fb7" class="amb nr fr be ns aid amc aie nw aig amd aih oa mz ame yd yg nd amf yh yk nh amg yl yo amh bj" data-selectable-paragraph="">3. 資料萃取</h2>
<p id="8c64" class="pw-post-body-paragraph mo mp fr mq b mr oo mt mu mv op mx my mz oq nb nc nd or nf ng nh os nj nk nl fk bj" data-selectable-paragraph="">傳統感測器採集的時序資料，無法區分哪些時段設備穩定運作、故障停機。需要依靠領域知識萃取穩定運轉的資料，以建立風機發電功率性能模型。</p>
<p id="c631" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">首先，下篩選條件，選取發電功率大於零的資料 (發電功率為零表示設備停機或無風)。</p>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu amk"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/1*r1MMiJzGBAYOp4QBbf9MMg.png 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/1*r1MMiJzGBAYOp4QBbf9MMg.png 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/1*r1MMiJzGBAYOp4QBbf9MMg.png 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/1*r1MMiJzGBAYOp4QBbf9MMg.png 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/1*r1MMiJzGBAYOp4QBbf9MMg.png 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/1*r1MMiJzGBAYOp4QBbf9MMg.png 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/1*r1MMiJzGBAYOp4QBbf9MMg.png 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/1*r1MMiJzGBAYOp4QBbf9MMg.png 640w, https://miro.medium.com/v2/resize:fit:720/1*r1MMiJzGBAYOp4QBbf9MMg.png 720w, https://miro.medium.com/v2/resize:fit:750/1*r1MMiJzGBAYOp4QBbf9MMg.png 750w, https://miro.medium.com/v2/resize:fit:786/1*r1MMiJzGBAYOp4QBbf9MMg.png 786w, https://miro.medium.com/v2/resize:fit:828/1*r1MMiJzGBAYOp4QBbf9MMg.png 828w, https://miro.medium.com/v2/resize:fit:1100/1*r1MMiJzGBAYOp4QBbf9MMg.png 1100w, https://miro.medium.com/v2/resize:fit:1400/1*r1MMiJzGBAYOp4QBbf9MMg.png 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/1*r1MMiJzGBAYOp4QBbf9MMg.png" alt="" width="700" height="393" /></picture></div>
</div>
</figure>
<p id="e656" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">其次，已知「高速端軸承溫度」與「風機風速」有關聯性，繪製雙變數的散佈圖，以拖曳圈選的方式，篩除與其他值表現明顯不同的離群值。</p>
<p id="39df" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">值得一提的是，所有編輯動作都會記錄在 Tukey 的右側欄。</p>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu aml"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/0*21rakfQXOOk3OWkp 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/0*21rakfQXOOk3OWkp 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/0*21rakfQXOOk3OWkp 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/0*21rakfQXOOk3OWkp 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/0*21rakfQXOOk3OWkp 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/0*21rakfQXOOk3OWkp 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/0*21rakfQXOOk3OWkp 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/0*21rakfQXOOk3OWkp 640w, https://miro.medium.com/v2/resize:fit:720/0*21rakfQXOOk3OWkp 720w, https://miro.medium.com/v2/resize:fit:750/0*21rakfQXOOk3OWkp 750w, https://miro.medium.com/v2/resize:fit:786/0*21rakfQXOOk3OWkp 786w, https://miro.medium.com/v2/resize:fit:828/0*21rakfQXOOk3OWkp 828w, https://miro.medium.com/v2/resize:fit:1100/0*21rakfQXOOk3OWkp 1100w, https://miro.medium.com/v2/resize:fit:1400/0*21rakfQXOOk3OWkp 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/0*21rakfQXOOk3OWkp" alt="" width="700" height="387" /></picture></div>
<div></div>
</div>
</figure>
<h2 id="9dca" class="amb nr fr be ns aid amc aie nw aig amd aih oa mz ame yd yg nd amf yh yk nh amg yl yo amh bj" data-selectable-paragraph="">4. 建立預測模型</h2>
<p id="118c" class="pw-post-body-paragraph mo mp fr mq b mr oo mt mu mv op mx my mz oq nb nc nd or nf ng nh os nj nk nl fk bj" data-selectable-paragraph="">已知發電功率（GRID_POWER）與風速（AMB_WINDSPEED）有相關性，先利用這個基礎背景知識建立一個初階模型。預測目標選擇「發電功率」（GRID_POWER），自變數選擇「風機風速」（AMB_WINDSPEED），演算法選擇最初階的廣義線性迴歸（GLM）。</p>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu amm"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/1*mMUEMFD7y4-asrZQoK28lw.png 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/1*mMUEMFD7y4-asrZQoK28lw.png 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/1*mMUEMFD7y4-asrZQoK28lw.png 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/1*mMUEMFD7y4-asrZQoK28lw.png 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/1*mMUEMFD7y4-asrZQoK28lw.png 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/1*mMUEMFD7y4-asrZQoK28lw.png 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/1*mMUEMFD7y4-asrZQoK28lw.png 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/1*mMUEMFD7y4-asrZQoK28lw.png 640w, https://miro.medium.com/v2/resize:fit:720/1*mMUEMFD7y4-asrZQoK28lw.png 720w, https://miro.medium.com/v2/resize:fit:750/1*mMUEMFD7y4-asrZQoK28lw.png 750w, https://miro.medium.com/v2/resize:fit:786/1*mMUEMFD7y4-asrZQoK28lw.png 786w, https://miro.medium.com/v2/resize:fit:828/1*mMUEMFD7y4-asrZQoK28lw.png 828w, https://miro.medium.com/v2/resize:fit:1100/1*mMUEMFD7y4-asrZQoK28lw.png 1100w, https://miro.medium.com/v2/resize:fit:1400/1*mMUEMFD7y4-asrZQoK28lw.png 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/1*mMUEMFD7y4-asrZQoK28lw.png" alt="" width="700" height="388" /></picture></div>
<div></div>
</div>
</figure>
<h2 id="e7bd" class="amb nr fr be ns aid amc aie nw aig amd aih oa mz ame yd yg nd amf yh yk nh amg yl yo amh bj" data-selectable-paragraph="">5.解讀預測結果</h2>
<p id="9690" class="pw-post-body-paragraph mo mp fr mq b mr oo mt mu mv op mx my mz oq nb nc nd or nf ng nh os nj nk nl fk bj" data-selectable-paragraph="">模型建立完成後，可以藉由RMSE、MAE、MAAPE 三個模型評估指標確認模型表現 (此三個數值愈小，代表模型表現愈好)。此外，Tukey 也提供視覺化圖形協助使用者評估模型表現。</p>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu amn"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/1*wFFFt1dPqqLim__6w3BbZg.png 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/1*wFFFt1dPqqLim__6w3BbZg.png 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/1*wFFFt1dPqqLim__6w3BbZg.png 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/1*wFFFt1dPqqLim__6w3BbZg.png 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/1*wFFFt1dPqqLim__6w3BbZg.png 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/1*wFFFt1dPqqLim__6w3BbZg.png 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/1*wFFFt1dPqqLim__6w3BbZg.png 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/1*wFFFt1dPqqLim__6w3BbZg.png 640w, https://miro.medium.com/v2/resize:fit:720/1*wFFFt1dPqqLim__6w3BbZg.png 720w, https://miro.medium.com/v2/resize:fit:750/1*wFFFt1dPqqLim__6w3BbZg.png 750w, https://miro.medium.com/v2/resize:fit:786/1*wFFFt1dPqqLim__6w3BbZg.png 786w, https://miro.medium.com/v2/resize:fit:828/1*wFFFt1dPqqLim__6w3BbZg.png 828w, https://miro.medium.com/v2/resize:fit:1100/1*wFFFt1dPqqLim__6w3BbZg.png 1100w, https://miro.medium.com/v2/resize:fit:1400/1*wFFFt1dPqqLim__6w3BbZg.png 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/1*wFFFt1dPqqLim__6w3BbZg.png" alt="" width="700" height="787" /></picture></div>
</div>
</figure>
<p id="26e2" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">以左上角的 Actual — Predicted Plot 為例，此為發電功率 (GRID_POWER, KW)實際值與預測值的對照圖。若預測完全正確，藍點會落在左下至右上的對角線上。分析時會特別查看誤差較大 (離45度線特別遠) 的區域。</p>
<p id="529c" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">觀察此圖，最明顯的地方是有許多發電功率實際值為 2000 (KW) 左右時，預測值的誤差都高估 (預測值為2000 ~ 2800)。</p>
<p id="0531" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">另一個比較不明顯的地方是，當發電功率實際值小於 2000 (KW) 時，圖形分佈呈 S 形，在實際值小於 1000 (KW) 時，預測值高估；在實際值介於 1000 至 2000 (KW) 時，預測值則低估。</p>
<p id="a66e" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">關於 Tukey 選擇這 RMSE、MAE、MAAPE 這三個模型指標以及這四種模型評估圖形的設計理念與進階判讀方式，未來會另起專文說明 (或是來<a class="af mn" href="https://aiacademy.tw/category/opening/" target="_blank" rel="noopener ugc nofollow">報名台灣人工智慧學校</a>，享受完整的手把手課程)。</p>
<h2 id="c2c3" class="amb nr fr be ns aid amc aie nw aig amd aih oa mz ame yd yg nd amf yh yk nh amg yl yo amh bj" data-selectable-paragraph="">6. 建立進階模型，進行模型比較</h2>
<p id="8bdf" class="pw-post-body-paragraph mo mp fr mq b mr oo mt mu mv op mx my mz oq nb nc nd or nf ng nh os nj nk nl fk bj" data-selectable-paragraph="">在觀察 baseline 模型 (GLM_MODEL_GRID_POWER) 評估頁面後，初步判斷發電功率 (GRID_POWER) 與風速 (AMB_WINDSPEED) 並非線性相關。故改用非線性的 GAM 演算法，建立進階模型。</p>
<p id="98b6" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">觀察「模型比較」頁面，會發現以 GAM 演算法建置的模型其 RMSE 明顯優於 GLM ，且 Actual — Predicted Plot 分佈也更貼近對角線。</p>
<p id="da8d" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">至此，便可將模型部署上線。</p>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu amj"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/1*zAIhZX2SBUU7XpXNlG2vkw.png 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/1*zAIhZX2SBUU7XpXNlG2vkw.png 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/1*zAIhZX2SBUU7XpXNlG2vkw.png 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/1*zAIhZX2SBUU7XpXNlG2vkw.png 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/1*zAIhZX2SBUU7XpXNlG2vkw.png 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/1*zAIhZX2SBUU7XpXNlG2vkw.png 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/1*zAIhZX2SBUU7XpXNlG2vkw.png 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/1*zAIhZX2SBUU7XpXNlG2vkw.png 640w, https://miro.medium.com/v2/resize:fit:720/1*zAIhZX2SBUU7XpXNlG2vkw.png 720w, https://miro.medium.com/v2/resize:fit:750/1*zAIhZX2SBUU7XpXNlG2vkw.png 750w, https://miro.medium.com/v2/resize:fit:786/1*zAIhZX2SBUU7XpXNlG2vkw.png 786w, https://miro.medium.com/v2/resize:fit:828/1*zAIhZX2SBUU7XpXNlG2vkw.png 828w, https://miro.medium.com/v2/resize:fit:1100/1*zAIhZX2SBUU7XpXNlG2vkw.png 1100w, https://miro.medium.com/v2/resize:fit:1400/1*zAIhZX2SBUU7XpXNlG2vkw.png 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/1*zAIhZX2SBUU7XpXNlG2vkw.png" alt="" width="700" height="389" /></picture></div>
</div>
</figure>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu amo"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/1*lsmLv1p7BaYGTON8lzuv4A.png 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/1*lsmLv1p7BaYGTON8lzuv4A.png 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/1*lsmLv1p7BaYGTON8lzuv4A.png 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/1*lsmLv1p7BaYGTON8lzuv4A.png 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/1*lsmLv1p7BaYGTON8lzuv4A.png 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/1*lsmLv1p7BaYGTON8lzuv4A.png 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/1*lsmLv1p7BaYGTON8lzuv4A.png 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/1*lsmLv1p7BaYGTON8lzuv4A.png 640w, https://miro.medium.com/v2/resize:fit:720/1*lsmLv1p7BaYGTON8lzuv4A.png 720w, https://miro.medium.com/v2/resize:fit:750/1*lsmLv1p7BaYGTON8lzuv4A.png 750w, https://miro.medium.com/v2/resize:fit:786/1*lsmLv1p7BaYGTON8lzuv4A.png 786w, https://miro.medium.com/v2/resize:fit:828/1*lsmLv1p7BaYGTON8lzuv4A.png 828w, https://miro.medium.com/v2/resize:fit:1100/1*lsmLv1p7BaYGTON8lzuv4A.png 1100w, https://miro.medium.com/v2/resize:fit:1400/1*lsmLv1p7BaYGTON8lzuv4A.png 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/1*lsmLv1p7BaYGTON8lzuv4A.png" alt="" width="700" height="519" /></picture></div>
<div></div>
</div>
</figure>
<h1 id="64cd" class="nq nr fr be ns nt nu nv nw nx ny nz oa ob oc od oe of og oh oi oj ok ol om on bj" data-selectable-paragraph="">風機性能監控看板</h1>
<p id="cee3" class="pw-post-body-paragraph mo mp fr mq b mr oo mt mu mv op mx my mz oq nb nc nd or nf ng nh os nj nk nl fk bj" data-selectable-paragraph="">在 Tukey 建立的 AI 模型可以以 API 的方式介接出來，以監控看板的方式呈現。圖表會跟著最新的資料即時更新顯示。當風機性能表現衰退至特定程度時，系統即會發送通知，提醒維修保養工程師安排停機保養</p>
</div>
</div>
<div class="mb">
<div class="ab ca">
<div class="amp vm amq vq amr ams ce amt cf amu ch bg">
<figure class="abv abw abx aby abz mb amw amx paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu amv"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/1*ABGGQFRFiL-p6SGjSs3WaQ.png 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/1*ABGGQFRFiL-p6SGjSs3WaQ.png 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/1*ABGGQFRFiL-p6SGjSs3WaQ.png 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/1*ABGGQFRFiL-p6SGjSs3WaQ.png 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/1*ABGGQFRFiL-p6SGjSs3WaQ.png 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/1*ABGGQFRFiL-p6SGjSs3WaQ.png 1100w, https://miro.medium.com/v2/resize:fit:2000/format:webp/1*ABGGQFRFiL-p6SGjSs3WaQ.png 2000w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 1000px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/1*ABGGQFRFiL-p6SGjSs3WaQ.png 640w, https://miro.medium.com/v2/resize:fit:720/1*ABGGQFRFiL-p6SGjSs3WaQ.png 720w, https://miro.medium.com/v2/resize:fit:750/1*ABGGQFRFiL-p6SGjSs3WaQ.png 750w, https://miro.medium.com/v2/resize:fit:786/1*ABGGQFRFiL-p6SGjSs3WaQ.png 786w, https://miro.medium.com/v2/resize:fit:828/1*ABGGQFRFiL-p6SGjSs3WaQ.png 828w, https://miro.medium.com/v2/resize:fit:1100/1*ABGGQFRFiL-p6SGjSs3WaQ.png 1100w, https://miro.medium.com/v2/resize:fit:2000/1*ABGGQFRFiL-p6SGjSs3WaQ.png 2000w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 1000px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:1000/1*ABGGQFRFiL-p6SGjSs3WaQ.png" alt="" width="1000" height="519" /></picture></div>
</div><figcaption class="mi mj mk lt lu ml mm be b bf z dw" data-selectable-paragraph="">風機性能監控看板</figcaption></figure>
</div>
</div>
</div>
<div class="ab ca">
<div class="ch bg ew ex ey ez">
<p id="7072" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">以下為各資訊欄目的詳細說明：</p>
<p id="2925" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">主畫面最上方的三個資訊欄位分別為即時風速、風機發電功率的健康程度、風機的齒輪箱健康程度。其中性能指標是由風機的即時資料和 Tukey 模型的預測值所計算出來的。</p>
<p id="8fd4" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">主畫面中間三個資訊欄位呈現風機的即時數據，包含：即時風速、即時發電功率與即時齒輪箱油溫。</p>
<p id="7f95" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">監控看板下方性能指標圖表，說明此風機當前即時的發電功率和齒輪箱油溫和 tukey 模型所繪製出的性能指標曲線的差異。</p>
<p id="1f93" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">以「發電功率性能指標曲線」為例，紫色的曲線是由 tukey 模型的試算結果所繪製出的「性能指標曲線」，其意義為：從過往的風機運轉經驗，在不同的風速時，所對應到的發電功率應該是多少。</p>
<p id="6090" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">綠色的點則是當前的即時風速與發電量所繪製而成。當即時數據的綠點離紫色的性能曲線其垂直距離愈近，表示風機性能狀態愈健康。</p>
<figure class="abv abw abx aby abz mb lt lu paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu amy"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/1*MdnqWVQrMoGoAp33A9NP4g.png 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/1*MdnqWVQrMoGoAp33A9NP4g.png 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/1*MdnqWVQrMoGoAp33A9NP4g.png 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/1*MdnqWVQrMoGoAp33A9NP4g.png 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/1*MdnqWVQrMoGoAp33A9NP4g.png 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/1*MdnqWVQrMoGoAp33A9NP4g.png 1100w, https://miro.medium.com/v2/resize:fit:1400/format:webp/1*MdnqWVQrMoGoAp33A9NP4g.png 1400w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/1*MdnqWVQrMoGoAp33A9NP4g.png 640w, https://miro.medium.com/v2/resize:fit:720/1*MdnqWVQrMoGoAp33A9NP4g.png 720w, https://miro.medium.com/v2/resize:fit:750/1*MdnqWVQrMoGoAp33A9NP4g.png 750w, https://miro.medium.com/v2/resize:fit:786/1*MdnqWVQrMoGoAp33A9NP4g.png 786w, https://miro.medium.com/v2/resize:fit:828/1*MdnqWVQrMoGoAp33A9NP4g.png 828w, https://miro.medium.com/v2/resize:fit:1100/1*MdnqWVQrMoGoAp33A9NP4g.png 1100w, https://miro.medium.com/v2/resize:fit:1400/1*MdnqWVQrMoGoAp33A9NP4g.png 1400w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 700px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:700/1*MdnqWVQrMoGoAp33A9NP4g.png" alt="" width="700" height="454" /></picture></div>
</div>
</figure>
<p id="f039" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">監控看板左側欄的參數設定，可以設定欲連線的 Tukey 主機與 Tukey 模型 API 金鑰。使用上，只需要輸入不同的 API 金鑰，即可調用不同的 tukey 模型。以上圖為例，GLM 模型和 GAM 模型的性能指標曲線完全不同，而該風機的機構設計是風速大於 15 m/s 時，發電功率即鎖定在 2000 kW/h，因此 GAM 模型較符合該風機的設計與運轉狀況，會優先選用該模型。</p>
</div>
</div>
<div class="mb">
<div class="ab ca">
<div class="amp vm amq vq amr ams ce amt cf amu ch bg">
<figure class="abv abw abx aby abz mb amw amx paragraph-image">
<div class="mc md ee me bg mf" tabindex="0" role="button">
<div class="lt lu amz"><picture><source srcset="https://miro.medium.com/v2/resize:fit:640/format:webp/1*PMOFtbOtyiC1JQS_gn2Sbg.png 640w, https://miro.medium.com/v2/resize:fit:720/format:webp/1*PMOFtbOtyiC1JQS_gn2Sbg.png 720w, https://miro.medium.com/v2/resize:fit:750/format:webp/1*PMOFtbOtyiC1JQS_gn2Sbg.png 750w, https://miro.medium.com/v2/resize:fit:786/format:webp/1*PMOFtbOtyiC1JQS_gn2Sbg.png 786w, https://miro.medium.com/v2/resize:fit:828/format:webp/1*PMOFtbOtyiC1JQS_gn2Sbg.png 828w, https://miro.medium.com/v2/resize:fit:1100/format:webp/1*PMOFtbOtyiC1JQS_gn2Sbg.png 1100w, https://miro.medium.com/v2/resize:fit:2000/format:webp/1*PMOFtbOtyiC1JQS_gn2Sbg.png 2000w" type="image/webp" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 1000px" /><source srcset="https://miro.medium.com/v2/resize:fit:640/1*PMOFtbOtyiC1JQS_gn2Sbg.png 640w, https://miro.medium.com/v2/resize:fit:720/1*PMOFtbOtyiC1JQS_gn2Sbg.png 720w, https://miro.medium.com/v2/resize:fit:750/1*PMOFtbOtyiC1JQS_gn2Sbg.png 750w, https://miro.medium.com/v2/resize:fit:786/1*PMOFtbOtyiC1JQS_gn2Sbg.png 786w, https://miro.medium.com/v2/resize:fit:828/1*PMOFtbOtyiC1JQS_gn2Sbg.png 828w, https://miro.medium.com/v2/resize:fit:1100/1*PMOFtbOtyiC1JQS_gn2Sbg.png 1100w, https://miro.medium.com/v2/resize:fit:2000/1*PMOFtbOtyiC1JQS_gn2Sbg.png 2000w" sizes="(min-resolution: 4dppx) and (max-width: 700px) 50vw, (-webkit-min-device-pixel-ratio: 4) and (max-width: 700px) 50vw, (min-resolution: 3dppx) and (max-width: 700px) 67vw, (-webkit-min-device-pixel-ratio: 3) and (max-width: 700px) 65vw, (min-resolution: 2.5dppx) and (max-width: 700px) 80vw, (-webkit-min-device-pixel-ratio: 2.5) and (max-width: 700px) 80vw, (min-resolution: 2dppx) and (max-width: 700px) 100vw, (-webkit-min-device-pixel-ratio: 2) and (max-width: 700px) 100vw, 1000px" data-testid="og" /><img decoding="async" loading="lazy" class="bg mg mh c" role="presentation" src="https://miro.medium.com/v2/resize:fit:1000/1*PMOFtbOtyiC1JQS_gn2Sbg.png" alt="" width="1000" height="576" /></picture></div>
<div></div>
</div>
</figure>
</div>
</div>
</div>
<div class="ab ca">
<div class="ch bg ew ex ey ez">
<h1 id="1b87" class="nq nr fr be ns nt nu nv nw nx ny nz oa ob oc od oe of og oh oi oj ok ol om on bj" data-selectable-paragraph="">結語</h1>
<p id="0d94" class="pw-post-body-paragraph mo mp fr mq b mr oo mt mu mv op mx my mz oq nb nc nd or nf ng nh os nj nk nl fk bj" data-selectable-paragraph="">技術供應鏈的創新，讓 AI 普及化近在眼前。本文完整示範如何使用 Tukey，讓設備保養人員在不用寫程式的情況下，輕鬆快速的建立風機設備性能監控 AI 模型，並部署到性能監控看板做設備即時監控。</p>
<p id="85bc" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">透過 Tukey，設備保養人員可以</p>
<ol class="">
<li id="e9d9" class="mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl ana alb alc bj" data-selectable-paragraph="">根據自己的專業的機構設計、風機運轉的知識背景，以資料視覺化的方式做資料清理。</li>
<li id="a7da" class="mo mp fr mq b mr ald mt mu mv ale mx my mz alf nb nc nd alg nf ng nh alh nj nk nl ana alb alc bj" data-selectable-paragraph="">利用簡易的操作流程，自行快速建立 AI 模型，評估哪個模型可以部署上線。</li>
<li id="be9f" class="mo mp fr mq b mr ald mt mu mv ale mx my mz alf nb nc nd alg nf ng nh alh nj nk nl ana alb alc bj" data-selectable-paragraph="">最後，將模型部署到性能監控看板，對風機的性能狀態做即時的監控。當風機性能表現衰退至特定程度時，系統即會發送通知，提醒設備保養工程師安排停機保養。</li>
</ol>
<p id="9a65" class="pw-post-body-paragraph mo mp fr mq b mr ms mt mu mv mw mx my mz na nb nc nd ne nf ng nh ni nj nk nl fk bj" data-selectable-paragraph="">如果你對 Tukey 產品，或是對風機性能監控看板的設計有興趣，歡迎<a class="af mn" href="http://chimes.ai/#contact-form" target="_blank" rel="noopener ugc nofollow">聯絡我們</a>索取更深入的介紹文件與範例影片。</p>
<p data-selectable-paragraph="">若對 Tukey 模型指標以及與進階判讀方式有興趣，請持續關注我們的系列文章，或者報名<a class="af mn" href="https://aiacademy.tw/category/opening/" target="_blank" rel="noopener ugc nofollow">台灣人工智慧學校課程</a>。</p>
</div>
</div><p>The post <a href="https://chimes.ai/2021/09/07/%e7%94%a8-tukey-%e6%89%93%e9%80%a0%e8%a8%ad%e5%82%99%e6%80%a7%e8%83%bd%e5%8d%b3%e6%99%82%e7%9b%a3%e6%8e%a7%e7%b3%bb%e7%b5%b1-%e4%bb%a5%e9%a2%a8%e5%8a%9b%e7%99%bc%e9%9b%bb%e6%a9%9f%e7%82%ba%e4%be%8b/">用 Tukey 打造設備性能即時監控系統，以風力發電機為例</a> first appeared on <a href="https://chimes.ai">Chimes AI</a>.</p>]]></content:encoded>
					
					<wfw:commentRss>https://chimes.ai/2021/09/07/%e7%94%a8-tukey-%e6%89%93%e9%80%a0%e8%a8%ad%e5%82%99%e6%80%a7%e8%83%bd%e5%8d%b3%e6%99%82%e7%9b%a3%e6%8e%a7%e7%b3%bb%e7%b5%b1-%e4%bb%a5%e9%a2%a8%e5%8a%9b%e7%99%bc%e9%9b%bb%e6%a9%9f%e7%82%ba%e4%be%8b/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
			</item>
	</channel>
</rss>
