Q: Could you elaborate on the key elements of a successful predictive model?
A: A successful predictive model would rely on:
Capturing CBS-related data across operations
- Defining failure events, collecting failure event sequence data (in Historian) and grouping it by manufacturers and duty types (such as Stackers/Reclaimers and Ship Loaders) for analysis
- Identifying variables that have an impact on CBSs, based on business, logistics and statistical importance
- Choosing the best-fit model using data-driven evidence
- Measuring the impact of variables on outcomes
- Generating reports on pro-active maintenance and interventions before failure
Here, the reference values for the predictors and the sample size for the observations are important to accurate forecasting.
Q: Which are the important predictive failure models?
A: There are two models that we should consider. The Belt Wear Out Rate Model and the Cox Proportional Hazards Regression Model.
In the Belt Wear Out Rate Model conveyor belts under observation may be grouped by duty types and studied for two options–
(a) Time-based Observation (b) Throughput-based Observation.
In Time-based Observation, the predictors being analyzed are observed over weeks and their impact on linear wear out rate of belt is denoted as the linear belt thickness wear out rate per week.
In Throughput-based Observation, the belts are observed for utilization and downtime. This is represented as the thickness wear out per million tons of ore.
It is seen from the predictive model analysis on CBS that key variables such as conveyor duty types and cycle time play a vital role in belt wear out rates.
The Cox Proportional Hazards Regression Model is based on the rationale that the instantaneous probability of failure of a machine is initially zero and increases cumulatively over time with usage. The hazards function describes the relative likelihood of a failure event occurring at time (t) denoted by f(t) conditional on the equipment having operated up to time (t).
The best prediction results are obtained when data provided by 2 CBS manufacturers don’t cross. There are no covariates influencing the relative likelihood of failure other than the explanatory predictors such as Belt Tension, Tonnage, Belt Type and Bearing Temp.