恭喜我們實驗室林家瑜、陳子定、彭浩然和王雪貞獲得2018第23屆大專校院資訊應用創新競賽亞太交流-英文組（獎金12萬）和產學合作組（獎金3萬）雙冠
Congratulations to Chia-Yu Lin, Tzu-Ting Chen, Hao-Jan Peng and Hsueh-Chen Wang on winning International ICT Innovative Services Awards 2018 in two group competitions (Asia-Pacific Exchange Group and Industry-School Cooperation Group) with prize 120,000 NTD and 30,000 NTD. Moreover, it is really exciting and encouraging that we won the first place from 12 teams in Asia-Pacific Exchange Group which including 8 teams from foreign universities.
Title of winning work: Quantized RNN Autoencoder for Time-Series Anomaly Detection
Well done! You all did a great job! Many thanks to the mentor of our team Prof. Li-Chun Wang. Besides, many thanks to our industry partner Tongtai (東台精機) and professors who gave supports and advices to our team, Prof. Hong-Han (Austin) Shuai and Prof. Jung-Hsien Chiang.
恭喜我們實驗室林家瑜、陳子定、彭浩然和王雪貞獲得2018第23屆大專校院資訊應用創新競賽亞太交流-英文組（獎金12萬）和產學合作組（獎金3萬）雙冠！ 值得一提的是，其中亞太交流-英文組是來自不同國家（澳洲、印尼、泰國、香港等）的競賽團隊，能夠從中脫穎而出讓你們的付出得到了最好的見證!
得獎作品名稱: Quantized RNN Autoencoder for Time-Series Anomaly Detection RNN及自動編碼器量化模型應用於工具機異常偵測。
感謝我們團隊及實驗室的導師王蒞君教授，老師辛苦啦! 同時，感謝我們的產業合作商東台精機以及給予我們團隊支持與指導的老師們，感謝帥宏瀚老師及蔣榮先老師！
More details about our winning work:
We propose QUAntized Recurrent neural network autoencoder for Time-series anomaly detection framework (QUART). The main idea of QUART is building a healthy model of machines. The new data are compared with the healthy model. If they are not matched, the anomaly is going to occur. The lack of run-to-failure data can be solved. QUART notifies anomaly 2hrs in advance, and the true alarm rate is up to 100 %. Besides, QUART is specially designed for data preprocessing. The training time can accelerate almost 120 times faster than the traditional model.
We cooperate with TongTai, which is the biggest machine tool company in Taiwan. To equip QUART with machine tools, the cost of repairing and failure products can be intensively decreased. More customers will prefer machine tools of TongTai. The revenue can be increased by about 60 million. QUART is not only for TongTai, but also for manufacturing industry in Taiwan. The APIs of QUART is developed. Companies can easily build maintenance models by APIs to save costs. The competitiveness of manufacturing industry in Taiwan can be significantly increased.
本專題開發出一個不需要完整機器壽命資料即可建置異常預測的模型框架Quantized Recurrent Neural Network Autoencoder for Time-series anomaly detection (QUART)，建立機台的健康模型，新資料與健康模型比對，即時預測是否即將發生異常要，QUART可提前兩小時進行異常預警，且預警率高達100%，同時QUART的框架針對資料的前處理做了特別的設計，讓整體模型在建置時的速度可以比傳統建置模型快120倍，透過QUART，模型有效率的被建置更新，且可協助工廠提早進行機台檢測作業。
本專題與機台設備廠商東台精機進行合作，對東台來說，將QUART導入工具機，可協助購買工具機的客戶降低定期維修、停工及製造不良品的成本，吸引更多客戶購買工具機，以其銷售的數控鑽床為例，一年的營收約12億，裝設預知保養系統的鑽床，可提升約5%的銷售率，約提升6千萬的營收。對製造業來說，QUART可協助其他製造業廠商低門檻的引入預知保養技術，降低成本，提升台灣製造業的競爭力。