Fault Detection and Identification on Pneumatic Production Machine

Intelligent Approaches for Anomaly Detection in Compressed Air Systems: A Systematic Review

“…As for machine-level research, the included publications focused on a production machine that has multiple pneumatically actuated components. Both Dobossy et al [51] and Rajashekarappa et al [52] focused on fault detection for a typical manufacturing production machine. Similarly, Cupek et al [53,54] and Demetgul et al [55][56][57] investigated anomaly detection techniques on mass-customised production systems involving material handling and sequential operations.…”

“…The study carried out by Dobossy et al [51] involved the use of a testbench, where twenty production process cycles were collected using a number of sensors, which were an accelerometer, airflow rate, pressure, and proximity sensors, a load cell, a microphone and a thermocouple. Data were collected for three reference processes in a separate manner.…”

“…These were a drilling operation, an assembly of press-fitting parts, and the transfer of components for an assembly operation. The dataset generation and collection were handled and managed on individual NI-cDAQ modules and controlled by the NI-cDAQ 9172 chassis [51]. Similarly, Kostkurkov et al [65] generated and collected two hundred operating cycles of data from an experimental setup.…”

“…A real-world dataset was used in [38], which was not generated by the research team, but obtained from the ECP braking system of a DK-2 locomotive from Shouhuang Railway. Within the manufacturing research field, Rajashekarappa et al [52] presented a real-world implementation, as opposed to the investigations carried out by [41][42][43]51,65]. Rajashekarappa et al [52] generated their dataset comprising airflow, temperature, and pressure sensors from a real-world pneumatic production system.…”

“…No faults were purposely induced during dataset generation, as a typical air compressor fault occurred during the data collection window [59]. On the other hand, the study performed by Holstein et al [60] involved a different approach from [59], but similar to those performed by [51,70]. This is because the dataset was generated by measurements collected from the structure-borne sound probe T10 from SONOTEC and a microphone MK301 from Microtech Gefell.…”