|
|
|
|
Big Data: Budget Development Support
Circuit Breaker Failure Rate Analysis data mining shows that the planned work done for any particular component type can vary dramatically from -year to year. One of the things not visible in that section was the actual labor that was required to accomplish the PM tasks documented in the work orders. Clearly, a goal for any organization planning to manage with information, especially information derived from data mining efforts, is to ensure not only are actual labor hours recorded, but that good estimates of labor hours are provided for planned work and that the planned number of PM work orders is close to the actual number of PM work orders implemented.
Currently in the utility community there are many CMMS or work management systems that have only been in use for a few years. The use of those systems is continuing to develop. Therefore, the quality and the content of the data change from year to year. This creates the need to assess the data as part of any decision support process, especially the budgeting process.
An example of this can be seen in a small work order data set for metal clad switchgear maintenance. There are 102 work orders in the data set representing the work done on 196 metal clad switchgear units over a period starting in 2001 and extending into 2003. Of the 102 work orders, 72 are PM work orders. Of the 72 PM work orders 55 have progressed to a status in the system at which they should have actual labor hours. Although this is a small population, it is still sufficient to demonstrate the use of data mining to monitor and improve a business process in support of an important business information need, that being good budgeting intelligence.
The data mining process began with the selection of the metal clad switchgear data records, both from the equipment asset catalog and from the work order history. The first provided the total count of the equipment, 196 units. The latter the work orders documenting the maintenance activity. Inspection of the PM program data indicates that there is a ten (10) year interval for the routine PM task leading us to expect about 19 to 20 PM work orders per year. As shown in Figure 5 with two complete years and a partial year for 2003 that the actual PM rate is running a little ahead of the planned rate.
Figure 5 Figure 5 shows the work orders per year, with the count of those with labor hours shown by the blue columns and those without labor hours shown by the dark red columns. The more important thing that can be seen is that there is a process in place that is improving the performance of the staff with respect to entering labor hours. Although 2003 is only a partial year of data, all of the records to date have labor hours. If the labor hours and the number of PM work orders expected to be done are completed, it should be possible to predict the staffing needed for this part of the maintenance process. This same process can be done for the whole PM program if it is properly reflected in the work management system. Those preparing the budget need to know if the actual hours generally agree with the estimated hours or not. Of the 55 work order records, 47 have labor hours. For those records with target completion dates from 2001 to 2003 one can see the over all quality of the labor hour estimates in Figure 6. - Figure 6 Figure 6 shows over the three years with data, 17% of the work orders do not have estimated hours and for 53% of the work orders the estimate is more than 20% low. The estimation is not accurate enough for these estimated hours to be used as the basis for labor budget estimates. Looking at the figures from year to year, in Figure 7, there may be some improvement in providing labor estimates. The continuing low labor hours estimates should prompt the responsible parties to determine the root cause. That cause may simply be that this data was not provided to the planners estimating these jobs. The cause may be related to the need to perform more "reactive" maintenance during this 10 year inspection than is documented in the maintenance guidelines. As is often the case, the data mining effort has produced something equally as valuable as an accurate answer. It has framed an important and focused question that can be answered. Moreover, the answer will help to improve the quality of equipment asset management.
Figure 7 The information found in figures 5, 6 and 7 could be generated, just as easily, for all PM tasks. This information could be provided to the responsible parts of the organization for review and action, such as improving the labor estimates for the various planned tasks. The other perspective may be to better scope the PM task activities to help the crews achieve the targeted estimated work hours. In either case the aggregate data could be provided with a variety of drill downs or break outs depending on the needs of the organization. The process of generating these outputs could be automated and scheduled for semi-annual review and even published to intranet web pages for use in planning and budgeting. The key to the effective data mining in the OLAP realm, again, is to keep drilling into the data to find the characteristics important to making decisions. The characteristics found by delving into the data may reflect real behaviors of equipment or people or they may reflect data anomalies. These exploratory activities in data mining require a combined set of skills. There is clearly a need for a person with the skills to access and manipulate data. Not a database administrator, but a good data analyst. The other knowledge base that is needed, which may be found in the same person or may require a team of people, is a strong understanding of the business and the equipment. An understanding of the work practices of the field crews, an understanding of te use of the CMMS and an understanding of the use and failure mechanisms of the equipment are necessary to maximize the return from data mining. This knowledge is needed to recognize a significant find, whether from OLAP efforts or multidimensional analysis.
Once the data capture and coding process has matured and the raw data quality has been improved it may be practical to apply the more exploratory multidimensional data mining tools and let them find the unknown or unsuspected relationships residing in the data. That effort requires a multidimensional data mining application and personnel with the specific skills to set it up and interpret the results. However, most companies will, as these examples indicate, accrue significant decision support benefits very early on their road to great quality data that will easily support full multidimensional data mining. Next: Backlog Analysis. Back to Abstract.
Back to Services.
|
|
|