Fault prognosis for Discrete Manufacturing Processes

Health Index Extraction Methods for Batch Processes in Semiconductor Manufacturing

“…Identifying the significant points (js, ks) by a method given in [19]. The set of S identified points is assigned S = {(1 js , 1 ks ), .…”

“…A survey of data-driven prognosis in SM presented in [1] shows that, almost the health indexes are calculated from the quadratic-form index such as squared prediction error (SPE), Hotelling's T-Square [9], Mahalanobis distance ( [13], [14]), k-nearest neighbor squared distance [15]. Meanwhile, [19] presents a new approach of fault prognosis where the health index extracted from the identified significant points of equipment, related to the degradation process.…”

“…In DR-E-HMPCA method, the degradation reconstruction-based identification is developed following the E-HMPCA-based monitoring proposed in [2], [17]. SP-PCA method presents the identification of significant points which are related to degradation [19]. The work's aim is to generate a raw health index for the discrete process manufacturing.…”

Energy efficiency in discrete-manufacturing systems: Insights, trends, and control strategies

“…Therefore, during the last years more robust control systems able to treat and use the available information have been developed. For example, currently, control systems oriented to prognosis and maintenance of manufacturing systems, based on the historical data, have been developed with the aim to predict and program the required changes or maintenance activities [107,108,109]. Additional to the prediction of maintenance tasks, strategies such as receding horizon control and advanced methods of process control (e.g., model predictive control (MPC)), have started to gain attention too, mainly, focusing on problems of energy efficiency and flexibility for planning and scheduling of processes at machine, line, and plant level [110,111,112,113].…”

Degradation modelling with operating mode changes

“…The health index can be generated by different approaches based on the available information on current and historical operating condition, such as: [2] where the HI is generated by combining the wavelet decomposition and isometric feature mapping techniques on raw signals of bearings; [5] and [6] extract the HI by identifying the significant points of sensors and observations of a batch process in semiconductor manufacturing. Once the HI is generated, the RUL is estimated using different methods, such as regression function ( [3], [4]), stochastic models ( [5], [7]), neural network ( [8], [9]), etc. Most of the models mentioned above do not take into account the changes of system operating modes due to different utilization contexts.…”