Smart manufacturing systems are considered the next generation of manufacturing applications. One important goal of the smart manufacturing system is to rapidly detect and anticipate failures to reduce maintenance cost and minimize machine downtime. This often boils down to detecting anomalies within the sensor data acquired from the system which has different characteristics with respect to the operating point of the environment or machines, such as, the RPM of the motor. In this paper, we analyze four datasets from sensors deployed in manufacturing testbeds. We detect the level of defect for each sensor data leveraging deep learning techniques. We also evaluate the performance of several traditional and ML-based forecasting models for predicting the time series of sensor data. We show that careful selection of training data by aggregating multiple predictive RPM values is beneficial. Then, considering the sparse data from one kind of sensor, we perform transfer learning from a high data rate sensor to perform defect type classification. We release our manufacturing database corpus (4 datasets) and codes for anomaly detection and defect type classification for the community to build on it. Taken together, we show that predictive failure classification can be achieved, paving the way for predictive maintenance.

Recent advances in sensor technology and computing power allow for the generation/utilization of larger and more diverse sets of data. These developments enable the creation of data-driven models that can support real-time decision making. Such a decision aid can allow for predictive maintenance (PdM) to be undertaken on a much greater scale in manufacturing plants. PdM includes data-driven prognostics and health management (PHM). To enhance the performance of a prognostic model, one key task is to collect high-quality data, and in the past this has often involved using a feature extraction method to get meaningful information from a large noisy dataset. However, such methods may not handle noisy data well or address measurement errors adequately. Consequently, extracted features may not represent a degradation process suitably as a machine approaches a failure or fault. Also, effects of sensor types on the feature extraction and prediction model have not been much explored yet. To overcome this limitation, dynamic feature extraction is proposed to mitigate the effect of noisy statistical features in a monotonic trend by introducing a statistical penalty. Then, the features extracted through the method are used to construct a health indicator (HI). With the available historical HI values, a probabilistic regression model may be used to forecast the time to failure (TTF) of rotating machinery with uncertainty propagation. To validate the proposed method, acceleration data were collected from rotating machinery for several run-to-failure cases. The proposed method is demonstrated to provide excellent forecasts of TTF for both accelerometer types.