Two theses, developed in collaboration with Vitesco Technologies and the Toulouse Institute of Artificial Intelligence (ANITI), focus on advanced maintenance diagnostics and extracting new knowledge from production data.
Deep insight into connected machines helps predict damage and failures and avoid unplanned downtime. When developing its system solutions and intelligent components for electric vehicles, the manufacturer is increasingly relying on preventive maintenance.
According to the company, “AI can improve decisions and automate some of them, such as manual visual inspection tasks typically performed by operators. The greater the number of operators performing manual inspection and the greater the variety of parts being inspected, the higher the potential error rate.”
Ultimately, for Vitesco Technologies, predictive maintenance “will save time on repetitive tasks and result in fewer errors due to the accuracy of the algorithm. This will allow employees to take on more complex tasks, leading to an increase in overall quality.”
PREDICTIVE MAINTENANCE: different approaches
To optimize equipment usage, predictive maintenance uses data collected by sensors and machine learning based on various approaches.
Continuous variable To estimate the remaining time of normal operation, we use the simulation of dynamic systems. Several possibilities: A continuous variable will be based on a failure that has already occurred in order to learn from it and prevent it from happening again. This is the so-called controlled approach: it uses machine data.
Anomaly detection Assessing the risk of a breakdown without analyzing any past events that have happened to the machine. This method evaluates the normality of the equipment. This is an unsupervised approach (without prior data).
Mixed approach This approach allows you to get information about the equipment through anomaly detection. For Natalie Barbosa Roa, it’s about asking the question, “There’s been a glitch, but what kind of glitch? The action to be taken is determined based on the detected fault (eg faulty transmission). To do this, you need to have the right information and assess where it is most critical, where losses are generated.”