INTRODUCTION
The MEC group focuses on research and development of data analytics and machine learning tools to enable root cause diagnosis, and fault-and-failure predictive models to predict machine failures, detect process anomalies and identify hidden capacities. Connectivity with machine controllers, data handling with various data protocols, analogue and digital sensors are also part of the competencies of the group.
RESEARCH THEMES
- Manufacturing Execution (MEX)
The MEX team divides its research into 2 major themes. The first theme has 2 areas of research focus to enable smart manufacturing in process anomaly detection and in shop-floor production optimisation. The first area of focus is to develop data mining and machine learning algorithms such as deep learning for correlation analysis to identify dominant contributing factors in early detection of process anomalies to enable quality prediction. The second area of focus is in development of advanced optimisation techniques for adaptive process control to address the challenge of low yield and quality issues in high mix low volume production environment. Research on multi-objective optimisation modelling is to develop bio-aspired machine learning algorithms to achieve optimal dispatching of work orders and mobile material handling vehicles for supporting just-in-time manufacturing.
The second research theme of research work by MEX team is in the development of localisation techniques with RFID and other wireless communications, to track workers, WIPs, finished goods, material so as to enable optimal dispatching of work orders and material handling systems. Activity recognition research targets at developing probabilistic-based model and classification for smartphone based activity recognition in order to reduce sensitivity to the variations in human activities. RFID and Bluetooth based localisation research is to develop autonomous algorithms for online, real-time and autonomous update of the localisation model.
- Shop-floor Health Management (SHM)
A major theme of work is in machine condition monitoring. This theme focuses on multivariate data modelling with data from add-on sensors and data from machine controllers, using a mix of joint time-frequency data and statistical time-series techniques, for fault detection and root cause analysis and manufacturing OEE analysis. A second area of focus is to develop methods to model minority datasets to address the data and class imbalance, which is a prevalent and challenging problem in fault diagnosis and prognosis. The third focus is in developing edge analytics, to overcome network latencies, to handle high speed sensor data.
A second major theme is in developing degradation prediction models, whereby parametric and semi-parametric models will be developed to handle prediction of multiple fault classes. The deteriorating condition of equipment as time progresses, when modelled accurately, can help users of equipment to make better and more reliable decisions to maintain inventory of spares, plan with more certainty on the commitment to keep production machines running. Prognostics approach include deep learning models, empirical mode decomposition techniques and Bayesian models to capture time-dependent degradation behaviour. The last focus of this theme will include the development of optimisation techniques of hyper-parameters in predictive models to enable prescriptive maintenance to give prescriptive advice on when and what needs to be repaired, bearing in mind the cost of lost production and cost of repairs.