Streaming analytics is both an art and a science of sophistication as it involves software engineering to implement machine learning for predictive analysis. In order to spread the usefulness of good streaming analytics across the backbone of our industrial sector, the mid-sized businesses, StreamAhead applies real-time performance monitoring and provides predictive adjustment to improve overall effectiveness of industrial manufacture and its component modules are scalable, cost-effective and easy-to-use. As more mid-sized businesses gain industrial efficiency benefiting from streaming analytics, their gross carbon footprint shrinks.
Underlying principle for StreamAhead’s inner workings-
StreamAhead’s fundamental thinking regarding how both good performance and low cost in two principles.
(a) Use of existing, proven open-source software as components for the construction of the streaming analytic system
(b) Development of precise modeling specific to the customized system application requirements
Modular components from open source-
As an example, in one of StreamAhead’s very first implementations in semiconductor packaging and testing industry, the system has three departments: Logs, Events, and Data. FIG. 1 outlines the components used for the construction of StreamAhead. In principle, the Logs department provides basic functionalities to select, cleanse, and govern the machine log in order to sort out data relationships from with the data flow. The process involves cleansing, extraction and normalization of unstructured and semi-structured data with a scalable parallel computing environment. In the Events department, the system processes data continuously and provides near-real-time processing with very low latency in tuning and monitoring manufacturing performance. The department further identifies anomalies and important events and issues alerts accordingly. In the Data department, the system fixes up its multivariate process speed predictions based on results obtained by combining models from all relevant stages of the packing and testing fabrication procedure. The prediction is then used as input to a high level model.
Precise modeling with relevant domain knowledge for target technology-
StreamAhead adopts a methodology that processes a system of incremental classification of data groups. See FIG. 2. The processing by this methodology is characterized by low, constant space and time complexity to implement instance-based learning and prediction, a modeling based on progressive learning to establish a converging scheme for producing suggestions for the facility’s production line operational parameter adjustments in a predictive manner. The model sorts out from relational factors (such as product item, material and recipe used for its production, among others) that are relevant to the production line performance. These factors are fed into the modeling for the convergence onto those true relevant fine adjustment suggestions.
Principal applications of StreamAhead-
Principal applications include semiconductor packaging and testing, semiconductor fabrication, printed circuit board production and li-ion battery production and so on. One best example is its application in a large IC packaging and testing service company. Up till the last tallied season, the company shed US$ 1.3 million annually in operations costs that include savings in equipment maintenance and related personnel expenditures and represents an estimated process efficiency improvement of about 3 ~ 6 percent.
ABOUT THE ENTRANT
Name:Cheng Juei Yu
Type of entry:teamTeam members:Yi Hsin Wu, Wei Wen Wu, Andy Peng, Julie Chen
Software used for this entry:Hadoop HDFS, Apache Storm, Redis, Maria DB, Open CPU