When entering a maintenance record into a CMMS, there's often a place where the operator can enter free-form comments. These comments may contain valuable information about the health of the equipment, any maintenance activities that were undertaken, and plans or recommendations for the future. The flexibility of the comments is attractive to operators, as a precise description of the observations can be recorded. However, using the comments information in data analysis usually requires some codifying of the comments, which is time-consuming and results in a loss of nuanced information. A machine learning approach to using comments data has been applied to predict the health of hydroelectric generating units. By embedding comments into a matrix to generate a "bag of words," and applying neural networks on the vocabulary, comments can be used to assess the current state of the asset and predict its next state. In this presentation, we'll discuss three machine learning algorithms in a way that's accessible and relevant to M&R practitioners: a classification method, a clustering method, and a neural network method. Each method will be partnered with direct applications to real-life maintenance problems, including lessons learned and potential uses in other contexts.
Dr. Janet Lam holds a PhD in industrial engineering from the University of Toronto. She has been working in the field of maintenance optimization since 2008, with an emphasis on optimal scheduling of inspections for condition-based maintenance. More recently, her research interests have extended to machine learning approaches for maintenance and asset management. Through her work at C-MORE, she has had the opportunity to apply academic research directly with industry partners, including those in mining, utilities, transportation, and the military. As the assistant director of C-MORE, Janet is involved with cultivating strong relationships with industry partners and developing maintenance engineering resources that are both useful and current.
