This paper presents an innovative methodology, from which an efficient system prototype is derived, for the algorithmic prediction of malfunctions of a generic industrial machine tool. It integrates physical devices and machinery with Text Mining technologies and allows the identification of anomalous behaviors, even of minimal entity, rarely perceived by other strategies in a machine tool. The system works without waiting for the end of the shift or the planned stop of the machine. Operationally, the system analyzes the log messages emitted by multiple data sources associated with a machine tool (such as different types of sensors and log files produced by part programs running on CNC or PLC) and deduces whether they can be inferred from them future machine malfunctions. In a preliminary offline phase, the system associates an alert level with each message and stores it in a data structure. At runtime, three algorithms guide the system: pre-processing, matching and analysis: Preprocessing, performed only once, builds the data structure; Matching, in which the system issues the alert level associated with the message; Analysis, which identifies possible future criticalities. It can also analyze an entire historical series of stored messages The algorithms have a linear execution time and are independent of the size of the data structure, which does not need to be sorted and therefore can be updated without any computational effort.
Industrial Machine Tool Component Surface Defect Dataset
