Real-time Machine State Monitoring using Sound and Machine Learning

Abstract

Machine operators frequently use sound as a source to determine the machine state. Machine sound is also widely used in various applications of operational state and diagnostic monitoring. Knowing the machine state from the sound can be considered as a human skill because experience operators will know much more about the machine state than a novice operator. Owing to innovations in digital smart technologies such IoT, big data, and artificial intelligence, many machine, systems, and processes are being digitalized and machine learning is utilized for process optimization, anomaly detection, predictive maintenance, and so on. If sound recognition of machine state as a human skill can be digitalized, these smart digital technologies can have significant impact on the manufacturing floors and can also enable survival and prosperity of small and medium sized manufacturing enterprises. In this talk, conversion from human skill to autonomy via smart acoustic sensing and virtual reality are proposed. A contact-based sound sensor has been developed and used to listen to machines as if an experienced operator is listening to the machine in order to monitor machine status and conditions using artificial intelligence. To enable sound stream automatically, the MTConnect framework was employed. The Log-Mel spectrum was adopted as a feature for convolutional neural network (CNN) model. In the CNN model training, a random search method was used to determine the hyperparameters of the CNN models. The machine learning model was used and implemented on a production machine to determine the machine state as well as operational data.