“Prediction,” goes an old Danish proverb, “is hazardous, especially about the future.” Smart Manufacturing systems have attempted for many years to utilize Artificial Intelligence to make predictions about how a system will be behave. Predictive maintenance systems attempt to make predictions about manufacturing hardware behavior, particularly unexpected behavior, though the use of data analytics. An advanced unsupervised machine learning method for the evaluation of time series sensor data is proposed for the purpose of predicting equipment failures in Chemical Vapor Deposition systems. Through the simultaneous evaluation of all available sensor data, the algorithm develops an expected model of the total process' behavior. The algorithm provides the responsible engineers with a simple scalar value that represents the deviation of a particular manufacturing batch from the expected behavior. Deviation from expected behavior may be interpreted as pointing to elevated risk of equipment failure.
Examples from recent Industrial Experiments in Wafer Fabrication CVD processes are shared to demonstrate the efficacy of the method.
.jpg "Wes Smith, MBA photo")