Design and validation of an IFDI System by the employment of Artificial Neural Networks

The new tendency of the manufacturing industry is: automating and monitoring all process activities. The automation, in particular in the mass-production, reduces passive times and allows a constant quality of the product. In fact, product quality comes out from the possibility to realize reliable measurements; consequently the relief and the diagnosis on the measurement systems become very important. In the future of the industrial automation, we will see the whole productive cycle without any operator control or supervision. In the modern establishments, automated systems, with diagnostics and control functions, are indispensable. In literature, we can find many I.F.D.I. techniques (Instrument Fault Detection Isolation), concerned with fault diagnosis in measurement systems. These techniques are characterized by different performances in terms of costs, intervention times and reliability. In the present work, the design and the validation of these systems concern the measurement station in a manufacturing company. The above company is operating in the sector of the superficial treatments of rolled metal. We have employed Artificial Neural Networks (ANNs). Exploiting the peculiar characteristics of ANNs, such as the elevated elaboration speed, the flexibility and the generality, an on-line diagnostic system has been conceived. After the company presentation, we have analyzed the challenges in the IFDI system design. Therefore, two ANNs have been planned: the first one to simulate the productive process and the second one to classify and generalize the on-line diagnostics.