Normal behavior modeling (NBM) is an approach to process, system, and equipment management and maintenance enabled by recent advances in industrial artificial intelligence (AI) and machine learning (ML). NBM is applicable to a wide range of operational fields-–in practice, any in which normal operation can be quantified from available data.
Seeing the big picture with industrial AI
There are several compelling advantages of NBM, accessed via an Industrial AI platform. The most significant benefit is the ability to continuously monitor and provide proactive alerts for maintenance problems on complex systems of components. These alerts are based not on individual parameters but a holistic understanding of the entire system. This is unique versus methodologies like statistical process control (SPC) and condition-based monitoring (CBM), which base their alerts on a single variable.
Normal behavior modeling requires no supervision
Normal behavior modeling can detect impending failures without having been previously exposed to those failure modes. This raises the notion of so-called supervised and unsupervised performance modeling. In supervised learning, the system is trained on historical data sets to recognize errors. A supervised system would learn about the various failure modes of a hydraulic pump by being shown all the ways a pump can fail along with data sets containing labels that correspond to these failure modes. The obvious weakness? An inability to identify unanticipated failure modes. The approach suffers, as well, from the need to manually identify and codify all potential failure scenarios.
In unsupervised learning, the Industrial AI platform is simply fed a large data set from the normally operating equipment. Once the model has learned what comprises “normal,” it will autonomously provide alerts on any situations deviating from this state. The model can continually revise its understanding of “normal.” Unsupervised learning is dynamic.
Industrial AI: Evaluation, scoring, and alerts
One of the key benefits of NBM is that it reduces many thousands of time-series data points to a single metric: a risk score. By deciding on upper and lower performance bounds, the model then will declare the system to be out-of-normal and generate an alert only when the risk score exceeds that upper criteria.
Through all of this, it’s easy to lose sight of one of the most important resources underlying the predictive maintenance process: the experience and expertise of technical staff members. They play key roles in making NBM development and use a success, including failure mode identification, the establishment of alert thresholds, and deciding when model retraining is needed.
Accessed via an Industrial AI platform, NBM techniques are useful in:
- Production equipment on oil platforms—Failure can mean millions in lost revenue as well as safety risks and environmental catastrophes. By AI modeling equipment temperatures, pressures, and other metrics, impending problems can be identified early, saving upstream operators millions of dollars and significant regulatory exposure.
- Manufacturing plants—Out-of-normal operations in manufacturing plants can result in safety hazards, environmental violations, and inferior product quality. Proactive identification using an Industrial AI platform helps to ensure profitable operations in frequently low-margin businesses.
- Commercial and military aviation—Jet engines and other complex airborne hardware are routinely subject to enormous operational stresses, and small problems can quickly cascade into expensive and dangerous situations, risking lives as well as the possible loss of immense capital investments.
Complex physical systems evolve with time. Equipment ages, maintenance occurs, tolerances change, and externalities like availability of time, people, money, and regulations fluctuate. As a consequence, our sense of what is “normal” for complex systems morphs; modeling needs the flexibility to adapt as circumstances evolve. Fortunately, NBM is uniquely well-positioned to respond to these inexorable changes, far more so than CBM, SPC, and other such methodologies.
To learn more about how SparkCognition uses NBM techniques to enable predictive/prescriptive maintenance, contact an Avathon Industrial AI expert.
