Algoritmi per l’analisi predittiva dei mal-funzionamenti di macchine industriali

In this paper, we describe an highly innovative methodological process for the efficient algorithmic prediction of possible malfunctions of indu-strial tool machines. Our proposal is based on efficient text analysis tech-nologies, which, within the huge amount of log messages produced by an industrial machine, look for anomalous patterns or behavioral deviations even minimal (but repeated or in association with other apparently devia-tions negligible) that could be indicators of possible future malfunctions. The system makes its predictions in real time, that is, without waiting for the end of shift, end of daily production or machine downtime. The solu-tion is efficient and optimal as it only needs to read each character of the log message text only once without further computational burdens of any kind. The algorithmic solution, once the danger of malfunction has been identified, can provide for "corrective" interventions that can be automati-cally applied in case of need, without having to wait for analysis and de-termination of the actions to be followed by personnel human, but possi-bly only asking for authorization to operate. The methodology is adaptable to any production domain, without cost-ly customization steps, but with a simple algorithm instruction step. Keywords: Analisi Predittiva, Malfunzionamenti, Text mining, algoritmi effi-cienti