Tool based on artificial neural networks to obtain cooling capacity of hermetic compressors through tests performed in production lines

Cooling capacity is typically used to control production quality of hermetic compressors for refrigeration, but such parameter is obtained under controlled conditions in laboratory. Due to time and cost involved, the quality control process considers samples of the production, however a simplified test compatible with cycle time is performed for each compressor in the line. This paper proposes an artificial neural system which can be integrated into the quality control of compressors in the production line to estimate their cooling capacity. A variation of a recently published method, which combines bootstrap techniques with Monte Carlo simulations, is used to assure the reliability of the results. The average difference observed between the proposed method and the results of regular tests done in laboratory was 0.65%, with standard deviation of 0.47%. The uncertainty of the estimates was 5.1%, which is close to the typical value observed in laboratory tests. The time needed to obtain both the inference of cooling capacity and the uncertainty associated to it is in the order of seconds. Results indicate that by integrating the proposed tool to compressor assembly lines it will be possible to estimate the cooling capacity parameter for all the produced units.

Automated summary

This paper proposes an artificial neural system to estimate the cooling capacity of compressors in the production line. The proposed method combines bootstrap techniques with Monte Carlo simulations to ensure reliable results. The average difference between the proposed method and laboratory tests was 0.65%, with a standard deviation of 0.47%. The uncertainty of the estimates was 5.1%, which is similar to the typical value observed in laboratory tests. By integrating this tool into compressor assembly lines, the cooling capacity parameter for all produced units can be estimated.