[4F4] Data simulator for real-time condition evaluation in manufacturing systems: evaluation in a test-bench
I Pietrangeli¹, G Mazzuto¹, A Gomez-Gonzalez², R Espadas², E Carrascal² and J Cuesta²
¹Università Politecnica delle Marche, Italy
²Ikerlan Technology Research Centre, Spain
Ensuring that a machine operates optimally is crucial for guaranteeing safety, efficiency, process stability and high-quality outcomes in manufacturing. Therefore, monitoring and evaluating machine behaviour to detect and, eventually, promptly correct any anomalies during operation is increasingly essential. In this context, a condition evaluation (CE) tool has been designed to extract a model of normal machine behaviour using various techniques, including artificial intelligence (AI) algorithms and statistical methods, based on cyclic system data. In this study, data were collected from a test-bench designed to evaluate bushing wear in a crank-slider mechanism. A Python-based real-time data simulator (RTDS) was developed to preprocess these data, resample or adjust signal speed as needed and transmit them in real time to a MongoDB (MDB) database. The CE tool reads the data from the MDB as if from a live process, enabling near real-time condition evaluation. The system was tested using two different algorithms: statistical and artificial intelligence (AI)-based (convolutional neural network (CNN)). Results show that while the statistical model offers more stable outputs, the CNN model demonstrates higher sensitivity to subtle system variations, making it more effective for detecting early-stage anomalies. However, training was limited to a reduced dataset due to hardware constraints, which may affect model generalisability and long-term reliability. Overall, the combined CE and RTDS framework enables dynamic, flexible and modular real-time condition evaluation, supporting predictive maintenance and decision-making processes in manufacturing.
