Experimental Applying Acoustic Emission to Fault Diagnosis and Prediction of Autonomous Devices

Industrial development is gradually transforming towards intelligent autonomy by the development trend of Industry 4.0. The mechanical system fault diagnosis by using prevention techniques is urgent and necessary. Thus, various abnormal diagnosis and prediction technologies based on AI (Artificial Intelligence) are extensively proposed in this paper. Moreover, it is using of Acoustic Information ML (Machine learning) systems to collect acoustic information, which can in-depth acknowledge system health and prevent system failures. The system is developed from acoustic data based on a data-driven ML system. By the way, it is including vibration signals and acoustic images gathered from machinery. This developed system uses a deep learning model to analyze and combine input acoustic feature data. Besides, there is a diagnosis model developed with AI learning methods that can be used for decision-making problems of various goals. The system can be widely used in many aspects, especially in monitoring machine status and product quality with a high degree of identification. The application of AI architecture plus the adaptation of ML scheme are employed to satisfy the requirements of the following practical operations. For example, the factory automation, error diagnosis and prediction of motor failure of automatic factory equipment, and even automatic feedback system after abnormal sound detection. Once the aforementioned scenario is combined with EC (edge computing) migration module can inspire innovative design concepts. Through the collaboration of practical technology, such as acoustics analysis, AI, EC, electromagnetics, communications, and other theoretical basis subjects, it is convenient for flexible changes made in response to the trend of Edge operation. Especially, it can be formed as a unique customized system. In addition, this paper investigates the embedded system in the application of smart speakers as a basis. Then it is jointing with audio recording (motor audio emission) to establish an ML model which is trained by the audio emission data. There a DSP system on the arm-4mf chip is adopted to complete the complex calculation of audio signal conversion digital, which is able to completely facilitate the judgment of audio signal emission from a specific motor. In this paper, the results from the build framework illustrate the accuracy of audio judgment can reach 85%, but the accuracy of judgment for motor audio emission still cannot reach 20% at the current stage. There are also many possible problems encountered in the research. Eventually, this paper provides an analysis method to accomplish the goal of solving judgment misalignment of the motor axis. The method is based on TinyML (Tiny machine learning) techniques so that the field of IoT can move toward the direction of smart energy saving. This article believes that the AIOT (AI Internet of Thing) in the future of AI popularization is bound to affect and change people’s lifestyles.