Demystifying False Positives and False Negatives for Predictive Maintenance

ARTICLE

False Positives and False Negatives are the two most important metrics for any Predictive Maintenance system

In this webinar, Mr. Sunil Vedula, Founder & CEO of Nanoprecise Sci Corp, talks about how Nanoprecise deals with the false positives & false negatives, and the strategy that Nanoprecise adopts, to keep these numbers low.

Mr. Vedula talks about how using output parameters is a comprehensive strategy that can differentiate between anomaly detection and fault detection. He does this by pointing out the similarities between machines and human body.

Following is an excerpt from the webinar:

False Negatives are the test results that indicate that a machine does not have a specific faulty component condition when it actually does have a faulty component. Having false negatives can be very dangerous for below reasons:

(i) It can cause mistrust on the solution.

(ii) It can cause massive monetary loss for the customer in terms of lost production or reputation and may affect the safety of personnel.

Due to these reasons, predictive maintenance industry has between 0 to 5% tolerance for the false negatives, especially if the equipment is semi-critical. If the equipment is supercritical, almost close to 0% false negatives are allowed.

A False Positive can be defined as a test result that indicates that a machine has a specific faulty component condition when the machine actually does not have a faulty component. Having false positives can be somewhat dangerous for below reasons:

(i) It can cause mistrust on the solution.

(ii) It can cause multiple site visits for the customer’s technician, which might just not be useful.

Due to these reasons, predictive maintenance industry has a tolerance between 0 to 5% for false positives, especially if the equipment is semi-critical. If the equipment is supercritical, almost close to 0% false positives are allowed.

The automated end-to-end system from Nanoprecise will notify the plant personnel about any weirdness or anomalous component in any of the sensor data like vibration, temperature, acoustic magnetic flux and humidity. It is important to note that an anomaly is not necessarily a fault, and therefore, plant staff need not reach to a conclusion that whenever we send an anomaly notification, it points to a fault. Therefore, in our notifications, one can notice only anomalies in the criticality of level 1, 2, and 3. Our AI ensures that the personnel are notified only after the fault has been properly identified through a unique fingerprint matching technique, which is why only faults of criticality level 4 and 5, are notified to the users. This is how Nanoprecise is able to keep the false positive and false negative rate low at the same time.

Watch the full webinar here: