Modification of the Kohonen Algorithm for Diagnosing Printed Circuit Assemblies

The article presents a method for thermal diagnostics of a printed circuit assembly (PCA) of a radio-electronic device (RED), based on the symbiosis of a computational and physical experiment, modeling thermal processes and machine learning technologies for data processing. In the course of solving the research problem, specialized software products were used, such as: computer-aided design (CAD) system the Solidworks, the circuit modeling package NI Multisim, engineering analysis and calculation tools Ansys. For the algorithmization of design procedures, the Python programming language was used. The results of computational experiments in the form of temperature values of electrical radio elements became the basis for the formation of a base of faults in the PCA. The fault base, in turn, was used as the input for training an artificial neural network built on self-organizing Kohonen maps. The result of the neural network’s function is a conclusion of the technical condition of the object being diagnosed.