Real-time preditive analytics on Overall Equipment Effectiveness (OEE)

Wong, Rui Zhen (2017) Real-time preditive analytics on Overall Equipment Effectiveness (OEE). Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

	Text (24 Pages) Real-time preditive analytics on Overall Equipment Effectiveness (OEE).pdf - Submitted Version Download (635kB)
	Text (Full Text) Real-time preditive analytics on Overall equipment effectiveness (OEE).pdf - Submitted Version Restricted to Registered users only Download (2MB)

Abstract

Overall equipment effectiveness (OEE) is a standard of performance measure on equipment productivity especially in manufacturing process. This metrics will helps the company to evaluate the performance of machines by identifying the underlying losses and operational effectiveness. Anomaly occurs when an instance deviates from a normal behaviour, in this case we meant when an OEE spike is being detected. A real time anomaly detection analytics is important for a business because some anomalies required immediate action, this will helps the management level for better and faster decision making especially on manufacturing scheduling. The methodology in this project involved 6 parts, preliminary studies, data preparation, attribute selection, model development, comparison analysis, and results validation. First, preliminary studies were done by reviewing literatures and self-questioning on the upcoming problems. Next, real industrial data is collected, compiled and undergo discretization to prepare the data for model training and testing process. After that, attribute selection was done by intuition, by expertise selection and wrapper methods such as Recursive Feature elimination (RFE) to reduce the dimensionality of dataset. In the model development process, Support Vector Regression (SVR), Linear Regression (LR) and Regression Tree (RT) classifiers were selected to develop a model for OEE value prediction and OEE spike prediction. The results shows that LR model by RFE selected attributes was performing best among others because it was able to achieved Root Mean Squared Error (RMSE) of 0.0013 and accuracy of 0.9892 when doing OEE value prediction and OEE spike classification respectively. In conclusion, the proposed model was able to predict OEE value and detect OEE spike (undesirable OEE drop), but the model can only tested with limited data (5 shifts) due to data constraints. Furthermore, this model is suggested to embed in smart manufacturing data analytics dashboard for operational monitoring purpose. Further work is suggested to collect actual real-time industrial data to test on the robustness of the proposed model. Last but not least, if the suitable dataset is successfully obtained, the next step is suggested propose a predictive analytic model where the model will be able to forecast future (one day ahead) OEE spike in real time, and provide actionable insights to notify user based on the causes of the spike occurrence.

Actions (login required)

View Item