Diagnosiss and Predictive Maintenance Using Machine

**Title**:Diagnosiss and predictive maintenance using Machine Learning (DoréMi)

**Project start date**: as soon as possible

**Contract duration**:12 months, possibly renewable once

**Keywords**:Diagnosis, Automated Production System, machine learning, LSTM, Case-based reasoning

**Project overview**

The DoréMi project focuses on diagnosing undesirable behaviors in Automated Production Systems (APS). The detection task involves identifying the system's operating mode, generating alerts based on predefined thresholds established through performance criteria, and quickly detecting any deviations from the system's nominal behavior. The diagnostic task, on the other hand, focuses on isolating and identifying the faulty component.

In this project, we propose leveraging one of the key tools of the future industrial landscape: the Digital Twin (DT). The DT concept involves digitizing a factory and virtually replicating its behavior. Many industrial solutions can match a machine's expected behavior with virtual commissioning. In this project, we will use DT to generate training data, specifically by injecting faults into the digitized system to produce large volumes of observations of faulty scenarios, all without risking real systems. We will utilize URCA's Cellflex 4.0 platform and associated JNs for data acquisition and experimental validation, while benefiting from the computing power of URCA's Romeo HPC platform.

In the DoréMi project, our goal is to develop an intelligent diagnostic solution that replaces traditional, often non-industrializable methods, with a new approach based on data learned from simulations of both normal and abnormal behaviors using a DT. This solution will provide human supervisory operators with real-time feedback on the system's health.

The project aims to test and enhance the solution proposed by Saddem and Baptiste (2022), which is a data-driven approach utilizing machine learning. The data will be supplied by the DT of the Cellflex 4.0 automated cell. Additionally, this project will compare the developed solution with that of Lang et al. (2021). Our objective is to establish a relevant benchmark dataset that considers the timing of fault occurrences. We will leverage URCA's Cellflex 4.0 platform and its associated Digital Twins for data acquisition and take advantage of URCA's Romeo HPC platform for experimental validation.

**Responsibilities**:The work consists in 1) improving and extending the exploratory results of on-line diagnosis of Automated Production Systems (SAP) controlled by Programmable Logic Controllers (PLC), based on the data in (Saddem and Baptiste, 2022), by optimizing the hyperparameters,2) comparing the diagnostic solution with that of (Lang et al, 2021) by setting up a relevant benchmark (dataset) taking into account the date of occurrence of faults using the digital twins of Celflex 4. 0, and 3) to publish the results by means of papers in one/two international conference(s) and to write a journal or review paper.

**Profile and skills required**
- PhD degree obtained after novembre 2021 (maximum 3 years before the date of recruitment).
- In-depth knowledge of machine learning and Python programming (NumPy, Pandas, TensorFlow, Scikit-learn libraries, etc.).
- Knowledge of automated production systems (especially discrete-event and hybrid systems).
- Ability to work in a team.
- Scientific rigor.
- Good oral and written English.

**PhD graduates from URCA are not eligible.**

**How to apply**

**References**

Saddem, R. and Batiste D., (2022) Machine learning-based approach for online fault Diagnosis of Discrete Event System. In: 16th International Workshop on Discrete Event Systems. Septembre 7-9, 2022. Prague, République Tchèque.

Lang, S., Plenk, V., Schmid, U. (2021) A Case-Based Reasoning Approach for a Decision Support System in Manufacturing. In: Fujita H., Selamat A., Lin J.CW., Ali M. (eds) Advances and Trends in Artificial Intelligence. From Theory to Practice. IEA/AIE 2021. Lecture Notes in Computer Science, vol 12799. Springer, Cham.

Status: Cadre

Pay: 2,300.00€ per month

Application Question(s):

- PhD certificate year after 2022

**Education**:

- Doctorate (required)

Work Location: In person

Expected Start Date: 02/12/2024