OMDEC - Optimal Maintenance Decisions Inc.

EXAKT, developed by the University of Toronto's CBM Lab (Condition-Based Maintenance Laboratory), is an advanced failure prediction software package. It was used to analyse the mill's pump failures, propose a new maintenance schedule designed to reduce the failures and to optimize the cost. EXAKT's statistical analyser applied to each of the pump's main failure modes prompted the CBM (Condition Based Monitoring) program to be retuned. Four key results were obtained:

• The Mean Time Between Replacement was reduced by 7%
• Retroactively applying this would have prevented 10 of 11 actual failures
• Resulting in a cost saving of over 30%
• The mill's Maintenance Engineers and Vibration Specialists approved both the EXAKT model and the approach - thus paving the way for its use elsewhere in the mill.

Mill management now have a software tool that will effectively predict equipment failure and the remaining useful life of its key assets.

Lessons Learned

1. collect and analyse only data that has relevance to actual failure modes.
2. combine failure probabilities with breakdown costs
3. when implementing an optimum replacement strategy based on pump condition, minor corrections in maintenance intervals, guided by CBM, yield significant change in cost and breakdown frequency.

Background and Objectives

This mill produces over 300,000 tons of Kraft Pulp each year - pulp that is destined for the converting mill and then on to market as facial tissues, paper towels and similar products. With the current acute stress on the market pricing for pulp and paper, bringing costs down and production up are key objectives for the mill's 400 employees.

Hence eliminating or substantially reducing the frequency of pump failure was clearly the key objective. However management was also seeking a way to balance the normal production pressure to keep running versus the cost of a failure. The missing link was a clear understanding of the probability of failure - this is EXAKT's task.

Typically in these situations, three types of failure consequence are recognised:

1. the cost of a breakdown repair (damage to the equipment, expedited parts, emergency and overtime crew costs etc)
2. the cost of lost production caused by the breakdown or slowdown
3. the loss of reputation as a consistent and reliable supplier, potential safety or environmental costs, potential penalty payments for non-supply for example

Methodology

The analysis started with a review of the mill's current data to evaluate its consistency, accuracy and adequacy for model building. The units being examined were Gould 3175L pumps which were used 24/7 at less than optimal rate in an attempt to reduce stress and prolong life. The core diagnosis was that turbulence was causing stress on the outer bearing.

33 bearing histories were examined in 8 pump locations, embracing 11 failures. For each of these, several measurements were analysed - five different vibration frequency bands, and the overall vibration reading.

Lesson learned #1: Of the seven streams of CBM data collected and analyzed, only two had any significance for failure prediction. The remainder are candidates for being discontinued.

To include the all-important event data, operating starts, out-of-service intervals and failure dates were extracted from the CMMS work history database.

From this data, a statistical model that would correlate the condition monitoring data with actual failure or potential failure events was determined.

An EXAKT Weibull Proportional Hazards model was developed to identify possible correlations between the vibration readings and the units' potential and functional failures for each key failure mode. Only two of the variables were proven to be predictors of failure. Much time and effort was henceforth avoided by no longer having to consider data that provides no additional information with regard to the probabilty of failure.

Lesson learned #3: a minor, but important correction in the maintenance interval yielded a significant change in costs and the frequency of breakdowns.

Using the company's estimate of the average ratio of 3.2:1 between breakdown cost and preventive replacement cost, combined with failure probability associated with each significant risk variable, the EXAKT decision model identified the optimum conditions for which the PM should be performed. This model balances the failure probability and the relative costs ofn prevention to breakdown. As each new set of inspection data is received, the model returns an optimal interpretation of the CBM data.

The EXAKT policy resulted in an average 7% reduction (from 571 to 529 days) in mean time to maintenance, as well as a marked shift from reactive to preventive maintenance. The new way of interpreting CBM data produced savings in the order of 30% for the chronic failure modes that had been plaguing these pumps.

The EXAKT prediction model answers the questions - Should the company keep the unit in operation until the next scheduled outage? Or should they take preventive action prior to the shutdown?

The EXAKT decision graphs above show the current status of the equipment. Each dot's value as plotted on the vertical axis indicates the weighted sum of significant CBM measurements. The horizontal axis measures the working age for a specific pump. As long as the current value is in the "Green Zone" ( the lower segment in Exhibit B), then the equipment can be expected to last until the next scheduled inspection. Readings in the "Danger Zone" (the upper segment in Exhibit A) indicates the company will lose money by continuing to operate - the equipment is overdue for breakdown. Maintenance managers now have the persuasive data to request an interruption of production, in order to undertake preventive maintenance, thereby avoiding the more dire consequences of breakdown.

The "Caution Zone" (between Green and Danger) indicates that the best decision will be to perform maintenance within the next inspection interval. Additionally an "expected remaining life estimate" is reported - important for scheduling maintenance on the unit. The formula at the foot of the screen calculates the weighted sum of significant monitored variables as determiend by EXAKT. The recommendations ("Replace Immediately" and "Don't Replace" respectively) are shown on the graph.

Data Adequacy

Data adequacy and data consistency are always an issue when building failure prediction models. Improving the quality of the prediction requires more relevant data and more consistent data. EXAKT's methodology calculates the statistical confidence levels built into the model; confidence levels less than 95% call for more data or more consistency in the data. This is achieved by focusing on the key variables, increasing the frequency of measurement and extending the period over which the measurements are taken.

The quality of data in a CBM program, without doubt, determines its effectiveness. In CBM, we must consider two types of data: 1. inspections, and 2. events. Inspection data is what we measure when we perform a CBM task, such as taking a vibration reading. Event data is what we observe when we perform maintenance or repair.

Inspection data must relate in some way to one or more failure modes of interest. The weaker that relationship (for example, due to noise or unaccounted for variables) the more event data we need to achieve confidence in the decision model's recommendations. EXAKT lets us know, through rigorious statistical tests, whether our CBM inspection data actually contains predictive capability. Thus EXAKT will guide us in designing effective CBM programs.

With the 95% confidence levels (ie, of no failure before the next outage), the operation can continue with the expectation of successful completion nineteen times out of twenty.

After introducing EXAKT into a maintenance program, new events and inspection data will continually augment our database. More data will povide better (more confident) decisions. EXAKT measures and reports on the confidence levels with which we interpret CBM inspection data. Hence, as the data acrues, our predictive models will be updated in a continuous, and measurable, improvement cycle.

Next Steps

With the success of this program, the mill management is plans to extend the application of EXAKT to other key equipment - with the intent of raising production and reducing costs even further. The company recognises that minor modifications need to be made to the information management system. These changes will ensure that the data will be readily available for analysis, thereby avoiding time consuming CMMS data cleansing processes.

The company is currently evaluating special software tools designed to assist this process - software that has been developed by the UofT's spin-off company OMDEC Inc, who are responsible for commercializing EXAKT.

Conclusions

1. Significant cost reductions were demonstrated - in the order of 30%
2. EXAKT failure prediction and decisions models were successfully developed and tested for the pump's key failure modes at the 95% confidence level
3. Some condition data were found to be of little or no value in predicting failure; their analysis and collection costs could be saved
4. Minor changes to the work order process wiill enable the mill's engineers to collect and extract key "as found" data easily for analysis and the generation of good predictive decision models;