Pillars of predictive maintenance: A data science perspective.
Predictive maintenance is essentially applying advanced data analytics to machine health management. The building blocks are.
1. Data science and AI expertise in predictive analysis
2. Domain knowledge about assets and machine operations
3. Underlying hardware and infrastructure support
Once the data is collected, cloud computing proves to be the most economic and efficient solution to train machine learning algorithms.
The performance of the machine learning models is evaluated through three measurements.
1. Accuracy
2. Precision
3. Recall
Using precision and recall as complementary metrics is a better approach. One of the most common complaints about machine learning is that it is too cautious and sends failure alerts when there is not a failure.
Precision is a good metric to measure how many fault predictions are correct. Recall focuses on correct predictions of actual faults, which works best when faults are rare but catastrophic to performance.
Are you thinking about improved labour productivity? Get in touch with us to find out more.
