BR-102024018131-A2 - Automated method for fault detection and diagnosis using machine learning.

Abstract

An automated method for detecting and diagnosing machine faults in real time, using machine learning (ML) resources that automates the entire process of model implementation, selection of the best model, and deployment for detection and diagnosis, so that a professional without any knowledge of machine learning can implement it.

Inventors

• GILBERTO DENIS DE SOUZA LEITE FILHO
• JOÃO FAUSTO LORENZATO DE OLIVEIRA
• ALDONSO MARTINS DE OLIVEIRA JUNIOR
• ALEXANDRE MAGNO ANDRADE MACIEL
• DIEGO JOSÉ RÁTIVA MILLÁN

Assignees

• MK DIGITAL LTDA
• FUNDAÇÃO UNIVERSIDADE DE PERNAMBUCO ? UPE
• GILBERTO DENIS DE SOUZA LEITE FILHO
• JOÃO FAUSTO LORENZATO DE OLIVEIRA
• ALDONSO MARTINS DE OLIVEIRA JUNIOR
• ALEXANDRE MAGNO ANDRADE MACIEL
• DIEGO JOSÉ RÁTIVA MILLÁN

Dates

Publication Date

20260317

Application Date

20240903

Claims (5)

1. 1. A METHOD FOR FAULT DETECTION AND DIAGNOSIS THROUGH MACHINE LEARNING, characterized by using, as input data, data available in industrial automation systems, such as time series, with values or states of digital and analog inputs and outputs;
2. 2. A method according to claim 1, characterized by not requiring the professional who will implement it to have knowledge of machine learning.
3. 3. A method, according to claim 1, characterized by requiring the professional who will implement it to have only information that is within the domain of people in the fields of automation or industrial maintenance;
4. 4. A method, according to claim 1, characterized by having all the processes of assembling the feature set, data preprocessing, implementation and training of machine learning models, selection of the model that presents the best performance, and execution of the model, without human intervention;
5. 5. A method, according to claim 1, characterized by considering the cyclical behavior of sequential machines, combining discrete events and continuous variables as input variables.

Description

Field of Invention [001] Automated method for detecting and diagnosing machine faults in real time using machine learning resources. It automates the entire process of model implementation, selection of the best model, and deployment for detection and diagnosis execution, so that a professional without any knowledge of machine learning can implement it. Fundamentals of the Invention [002] Several studies have focused on developing accurate ML models to reduce downtime in production and improve process quality; most deal with continuous processes. Recently, Kojuk et al. [1,2] developed a method and approach to build a decision support tool that combines supervised and semi-supervised techniques to detect and diagnose faults performed on continuous process data. Ren et al. [3] developed a methodology based on deep belief networks and several models to perform fault detection for complex systems. Chiu et al. [4] proposed a method using random forest, time series and long-short term memory (LSTM) to achieve faster monitoring and corrective adjustment of machines. Furukawa et al. [5] used the change score generated by ChangeFinder as new features in the SVM to classify and anomalous conditions improving detection speed and accuracy compared to the original SVM. Finally, Makridis et al. [6] proposed a method that combines a set to perform the task of predicting failures in maritime vessels. [003] Regarding fault detection and diagnosis tasks in discrete event systems, such as industrial sequential machines, most studies represent system behavior using Petri nets or finite state machines to implement diagnostic approaches. For example, Cohen et al. [7] developed a hybrid approach that uses Petri nets to guide fault diagnosis driven by PLC (Programmable Logic Controller) timed cyclic event systems with 97.2% validation accuracy. Furthermore, Lee and Chuang [8] developed a Petri Net-Based Fault Diagnosis System for Industrial Processes. Their solution involves learning the normal machine behavior, designing a Petri net from it, and implementing PLC routines based on logical combinations that detect whether the machine behavior is normal or anomalous. Finally, Ghosh et al. [9] developed an automated fault detection tool for PLC-controlled manufacturing systems. Their approach focuses on learning the states of a sequential machine over time to detect when a sensor or actuator state change occurs at an unexpected moment. [004] Regarding automated approaches or methods based on machine learning (AutoMLs) for fault detection and diagnosis, as is the case with the method presented in this document, Larocque-Villiers et al. [10] and Li et al. [11] developed AutoMLs for intelligent fault detection in bearings and gearboxes, Petition 870250039713, dated 05/15/2025, page 5/13 r y ’ respectively. Kefalas et al. [12] investigated the use of AutoML for estimating the remaining useful life of aircraft engines, and Nascimento et al. [11] studied the diagnosis of operating conditions and sensor failures using an AutoML. In all cases, AutoML proved to be very efficient and save significant time. [005] In the state of the art, there are no solutions involving machine learning for fault detection and diagnosis in discrete machines. The solutions found that employ machine learning for fault detection and diagnosis are suitable for continuous processes, while those dealing with discrete event systems, such as discrete machines and cyclic sequential machines, use Petri nets or State Machines. [006] Furthermore, although continuous variables can indicate an impending failure and contribute to improving the fault detection and diagnosis task, the combination of continuous variables with discrete events does not exist in the state of the art of fault detection and diagnosis in discrete machines. For example, an unusual temperature in a specific device may precede its damage. In addition, the typical sequential cyclic behavior of discrete machines has not been incorporated into the state of the art of machine learning to perform the fault detection and diagnosis task. [007] Finally, even if the state of the art had alternative solutions involving machine learning for fault detection and diagnosis in discrete machines, another challenge would still need to be faced: the current industrial workforce practically does not include professionals ready to use machine learning, such as data scientists [13]. In view of this, they suggest exploring automated machine learning (AutoML) resources [14-17]. These allow professionals without any machine learning knowledge to implement such solutions; however, none of the identified solutions are capable of performing fault detection and diagnosis in discrete machines. Brief Description of the Designs Figure 1 Demonstration of automation of all Machine Learning (ML) processes. Figure 2 Generic example of the proposed feature set assembly (discrete events that occur in a machine cycle). Figure 3 shows that 70