Bringing predictive maintenance to the factory floor
The modern-day factory is a complex, often high-tech environment. Assembly lines comprise a range of equipment and components, each of which performs a specific task in a given order. An issue with just one component can throw a wrench into the works, bringing the entire manufacturing process to a screeching halt. Such stoppages are extremely expensive, so factories place a significant emphasis – and budget – on maintenance. The problem is that most maintenance activities are either routine, meaning they happen regardless of whether a piece of equipment needs fixing, or reactive, meaning they happen after something breaks down. Although both approaches help reduce the risk of a lengthy shutdown, neither does a very good job at preventing the shutdown in the first place. What factories need is predictive maintenance – which is exactly what the EU-funded PROGRAMS (PROGnostics based Reliability Analysis for Maintenance Scheduling) project aims to provide. “We intend to extract information from every factory level – controllers and sensor data, maintenance reports, operator experience, physical characteristics, etc. – and, using an artificial intelligence-based algorithm, determine the optimal scheduling of maintenance activities,” says project coordinator Sotiris Makris, who heads the Robotics, Automation and Virtual Reality unit of the University of Patras Laboratory for Manufacturing Systems and Automation in Greece. “By minimising the impact that maintenance activities have on the production plan, we can help increase productivity and decrease costs.”
A game changer in predictive maintenance
The project’s defining outcome is the development of an innovative tool for predicting a part’s remaining useful life (RUL). “Using smart algorithms that exploit AI-based models and data collected from the field, this tool is nothing short of a game changer in predictive maintenance,” notes Makris. According to Makris, knowing a part’s RUL allows a company to make short- and long-term capital expense calculations. “Such information supports decision-making on ordering spare parts or scheduling maintenance activities,” he adds. “It can also be used to ensure that personnel are properly trained on performing a specific maintenance task.” In developing the tool, researchers faced a unique challenge. “Surprisingly, no breakdowns occurred during the project period, meaning we didn’t have any data on breakdown cases to feed into our AI algorithm,” explains Makris. Instead, the project used AI-based models to simulate the underperforming status of the equipment. “Machinery breakdown is preceded by some deterioration in performance and a resulting decrease in product quality,” adds Makris. “This adjustment was actually very welcomed by our industrial partners as it allows them to avoid not only a breakdown, but also the preceding decrease in performance.”
An important milestone
The project is now working to advance the maturity of its tool and move it towards commercialisation. “Our ultimate goal is to prepare robust, AI-based predictive maintenance solutions that can be integrated into industrial applications,” notes Makris. Achieving this goal has been made easier thanks in part to the project being a member of the ForeSee Cluster, a network of six EU-funded projects working on predictive maintenance technologies. “Our participation in the cluster has been an important milestone for the project,” concludes Makris. “Not only does it ensure that our results are incorporated into standardisation efforts, it also introduces us to an array of stakeholders who can use our work as a foundation for their own research.”
Keywords
PROGRAMS, predictive maintenance, maintenance, factory, smart algorithms, artificial intelligence, AI, data analysis, remaining useful life, assembly lines, manufacturing, industrial applications