Precise pre-delivery quality management of dairy products based on consumer feedback driven deep learning and knowledge graph

The quality of dairy products is of great significance to people's health. Many dairy enterprises cost a lot on the sampling inspection at the time of product delivery. If the company adopts mass sampling, the inspection cost will be very high. It is expected to reduce the sampling sample size as much as possible on the premise of ensuring that the problem product is detected. In this research, we propose the solution of precise detection by using consumer complaints as the intuitive feedback of product quality and make quality prediction before product delivery. First, monitoring information along the production line and consumer complaints information are collected and associated according to chronological orders. Secondly, neural networks are used to predict problematic products and to classify the types of quality issues by learning monitoring-compliant pairings. Finally, a specialized knowledge graph interface is developed to facilitate the operation of abnormal information entering and the obtention of prediction results. In practice, the prediction accuracy of product quality problem before delivery attains 92.7% and 90.4% for two kinds of milk, SIG and Tetra, which surpasses sampling inspection by over 10%. Besides, the knowledge graph interface shortens the quality problem record time by 34.5% and reduces the misinform rate by 57.1%. This work helps dairy engineers to discover potential problematic products before delivery, thus making the pre-delivery inspection more precise and reducing the quality management cost for dairy enterprises.