Automated preventive maintenance model reduced irrelevant fault alerts by 70%
Developed an automatic preventive maintenance framework for over 800 sensors across customer’s machines to detect faults and reduce scheduled maintenance costs
A leader in industrial automation solutions wanted to check the health of their hydroforming presses installed across geographies and provide timely maintenance to those that ran a risk of breaking down to avoid downtime for their consumers. The customer had collected over two years of data from each hydroforming press from over 800 sensors. With this vast amount of data from sensors, it became challenging to identify the sensors or combination of sensors that contained signals of machine failures. Their existing machine learning system was unable to detect new failure patterns and was not scalable through different projects.
Sigmoid built a machine learning solution to automate the detection of behavior patterns of different equipment as observed across all the sensors. The solution could automatically, in an unsupervised manner, build a pattern library for individual sensors displaying common vs rare failure patterns. The common patterns included historical failures identified as risks based on previous maintenance history while rare patterns were potential new failures that emerged from behaviors for which the sensors were not trained. The system could identify emerging patterns from data flowing in real time for preventive alerts. We predicted risk events and generated alerts based on the health state of the machine and predicted the probability of risk events to reduce unscheduled maintenance and downtime.
Sigmoid’s preventive maintenance algorithms were easily scaled on the customer’s entire available data set that resulted in significant savings in infrastructure cost of computing and storage.
data size per month
reduction in irrelevant alerts
sensors streaming data