IMPLEMENTATION OF INTELLIGENT QUALITY CONTROL SYSTEMS AT FLOUR MILLS IN UKRAINE

Abstract

The article is devoted to the urgent problem of modernization of the agro-industrial complex through the transition to automated monitoring of product quality in real time. The author justifies the need to abandon traditional laboratory methods, which have significant time delays (from 2 to 4 hours), which creates risks of producing defective products in the event of technological failures. The proposed solution is based on the development of a cyber-physical system (CPS) based on the Edge Computing architecture. This allows you to transfer the decision-making process directly to the production line, eliminating delays in data transmission to the cloud and eliminating the impact of electromagnetic interference typical of industrial zones. Special attention is paid to safety: the system hardware is designed taking into account the explosive hazard of flour dust (zones 20–22 according to the ATEX classification), which requires the use of sealed housings of the IP65/IP67 standard and limiting the surface temperature of the devices. The technical implementation includes the use of industrial cameras with a global shutter (Global Shutter), which prevent distortion of the image of the moving flour flow. For quality analysis, the MobileNetV2 neural network architecture is used, optimized using TensorRT INT8 quantization, which allows achieving classification accuracy of over 98% with minimal computational costs. The mathematical model of the system is based on the analysis of the CIE Lab* color space and hybrid processing of visual and parametric data from sensors. The scientific novelty of the work lies in the implementation of a dual approach to analysis: in parallel with the classification of varieties, an algorithm based on unsupervised learning (autoencoders) works. This allows you to detect previously unknown types of defects or foreign impurities (insects, metal particles, etc.) by analyzing deviations from the mathematical model of the “ideal product”. The proposed system provides stable operation with a response delay within 15–45 ms, which is critically important for the automatic operation of the defect cutters. The implementation of such a CFS contributes to the harmonization of Ukrainian standards with EU requirements for food safety.