Enhancing industrial quality control efficiency: an innovative deep learning approach for sustainable process monitoring

A major concern in traditional industrial monitoring is the strong environmental impact, mainly related to inefficiency of classic paradigms. In fact, typically monitoring systems rely on the presence of human operators responsible for the detection of errors or faults. However, this activity is heavily influenced by many factors like subjectivity or physical conditions (e.g., fatigue, lighting), making this strategy ineffective in terms of costs (both environmental and company-wide) and results. For instance, when the process involves the control of production lots, if the operator identifies any anomalies the whole batch is discarded. Sustainability and performance can be achieved by the automation of the monitoring process. In this regard, we propose an innovative method based on a deep neural network that can discriminate between correct and faulty items in a production batch. Our model allows to significantly reduce disposal costs, since it analyzes each item rather than considering the whole batch, thus preventing the waste of potentially usable resources. Furthermore, the methodology enables the optimization of the monitoring quality and lightens the responsibilities of the human operator, who only reviews the model outputs and generates relevant statistics for the company. We provide a thorough description of the proposed model in the context of the monitoring of transparent tubes within the production process of a company dealing with plastic consumables. Preliminary experiments we have performed on a real dataset confirm the effectiveness of the proposed method.