An LSTM-Based Approach for Anomaly Detection in Bolt Thread Rolling Process

Bolted fasteners are used in machine tools, bridges, transportation machinery, power plants and other mechanical structures. The most common use of such elements is in the automotive industry, where there are thousands of threaded connections in the assembly of a passenger car. In the production of these elements, the thread rolling method is used for threading and this method is used in 90% of threaded connections. A wide variety of defects occur during threading operations. The cost to the industry due to lost time and materials is directly related to the percentage of rejected products, making it crucial to find an effective way to address these defects. In this paper, a method using a long short-term memory (LSTM) approach is proposed to detect defects in the thread rolling process that lead to defective bolt production. The proposed approach collects time series data from the machine with a strain sensor during the thread rolling process. Then, the obtained signals are taught by an automatic LSTM-based encoder for the robust case and anomalies occurring in defective cases are detected.