Predictive Analytics Service

Main articles:

• Predicative analytics (predictive, expected, predictive) Predictive analytics
• Machine Learning

Service of predictive analytics is intended for detection of anomalies (deviations) in indications of sensors on the equipment by means of artificial intelligence (AI) with application of methods of machine learning (Machine Learning).

2021: Service Description

According to information as of November 2021, the predictive analytics service allows you to determine anomalies in the operation of equipment from the readings of sensors and automatically classify them into developing types of failure. Service operation is based on mutual prediction of sensor values at each time point. The model is trained, remembers the "normal" mode of operation and then is able to "predict" when any sensors behave "not as expected from them." Result of service operation - report on anomalies (deviations) in equipment operation. The anomalies found are classified according to a refillable diagnostic database and indicate possible developing types of failure that can be considered in the RCM analysis.

In the algorithm for determining anomalies, the service uses a multi-regression model (a model for predicting the continuous values ​ ​ of all sensors) based on the machine learning algorithm - gradient boosting over decisive trees. Sensor readings at each time point are predicted based on other sensor readings.

Integrations: Predictive Analytics integrates with RCM class systems (for example, "1C: RCM Reliability Management").

Target users: Reliability specialists.

User Experience

• The new source of possible failure modes improves the completeness and reliability of the RCM analysis process.
• Possibility of early detection of developing latent failures in equipment.
• Accumulation of expertise about equipment in the system in the form of diagnostics.