Machine-Learning based Predictive Maintenance for industrial components or plants
Wait / Service before something happens
With Katulu’s Predictive Maintenance solution, you can use real-time data to predict the likelihood of impending failures or maintenance needs. To achieve this, Katulu relies on machine learning models, trained with data series of many hundreds to thousands of classified plant states.
Your advantages are manifold: By avoiding unplanned breakdowns, your plant availability is increased (OEE). Sensor-based early detection of maintenance requirements helps to extend the lifetime of your assets. Additionally, you can significantly improve your plant utilization through optimization based on real-time status information.
The transition to machine learning based predictions requires initial effort (data collection, data scrubbing). Data scrubbing includes quality checks of the collected raw data, duplicate cleansing, data standardization, data normalization and data enrichment. Depending on the quality of the data, this step can account for up to 80% of the development of a machine learning model. The use of OPC-UA can significantly increase data quality, which in turn significantly reduces the data scrubbing effort.
In machine learning modeling, a model is selected/developed, trained, evaluated and tested on the basis of the processed data. Since there are already applicable models for many problems, in about 90% of the cases the selection and application of the right model is decisive. The development of a new model is only necessary in about 10% of the cases.
Our Machine Learning based Predictive Maintenance solution is based on our Condition Monitoring solution as a foundation, to enable a cost efficient implementation in successive stages of development.