Instead of the value-based analyses use by most condition-monitoring systems, the new system uses a model-based approach. Rexroth says that this advances the process from state monitoring to truly predictive analysis and data-driven, anticipatory maintenance.
During an initial phase, a machine-learning algorithm determines the healthy state for each component, based on sensor readings such as pressure, vibration, temperature, and oil quality. This phase may only last a few days if the installation carries out the same functions under similar conditions all the time. But if it is used only occasionally, or in different ways to manufacture different products, the phase may last longer.
ODiN then uses its data-based model to define a “health index” for each component. If a measured value deviates only temporarily from the tolerance range, an error warning may not be generated, because wear-and-tear can seldom be detected from one signal. However, if the health index deteriorates based on changes detected by multiple sensors, then the system warns of a problem.
“Diagnosing wear-and-tear in industrial applications is an extremely complex task,” says Rexroth project manager, Tapio Torikka. “Statistically, there is only a 13% probability of an issue being detected by chance, while an expert monitoring the system by traditional means has a 43% chance of detecting it. Our system has a detection rate of 99%.
“The system acquires all of the necessary information from the sensor data and machine-learning methods, then converts this into knowledge,” he adds. “The health index therefore not only shows the state of the assembly currently being monitored, but also gradual changes to upstream and downstream mechanical or hydraulic systems. If movements take longer or require more power, this indicates wear-and-tear. ODiN gives corresponding instructions in its regular health index reports and helps to create specific recommendations for action.”
“Even ODiN cannot fully eliminate the risk of plant downtime,” Torikka concedes, “but we can reduce the risk so significantly that the costs for the system are generally already recouped after the first prevented downtime.”