Similarity Based Methodology for Industrial Signal Recovery

The tremendous amount of data generated in the industry provides a massive opportunity to mine that data for decisions, such as prediction of outgoing product quality, process monitoring, etc. In addition, unlike computer and social networks, in the industrial data, the information is not directly observable and is embedded in the signals emitted during the corresponding processes, etc. However, in many cases and for many reasons these sensor signatures are not properly received at the very source causing missing segments in the signal sets. On the other hand, in many manufacturing facilities, large amounts of historical records of past sensor readings are available and can be used to enhance and reinforce the signal recovery process. In this paper, we propose the so-called match matrix methodology which uses signal similarity metrics to regenerate the missing segments in a signal from historical signal records. Three different incomplete signal set situations are simulated using a large dataset from a modern semiconductor manufacturing fab. The proposed method is validated utilizing the dataset and the results demonstrated a high fidelity in signal recovery in the all three cases.