Case Study: New York Air Brakes

Gerald DeFoe, a Six Sigma Black Belt in New York Air Brake’s Quality department, oversees Six Sigma implementation, training, and nontransactional projects at New York Air Brake. As a Black Belt, DeFoe relies on Crystal Ball to ensure that part production and repair processes are as cost and time efficient as possible.

DeFoe was recently asked to examine the company’s replacement process for a multiassembly locomotive braking product called CCBII (Computer Controlled Brake). Each of the CCBII subassemblies is composed of a mechanical portion and an electrical portion. Company policy states that if a part is returned due to a pure electrical failure within 120 days of its manufacturing date, then the electrical part will be repaired or replaced. The mechanical portion is used as is as a replacement unit on another locomotive. The liability of this practice is that the mechanical part could end up on a locomotive younger than the replacement, and NYAB wanted DeFoe to determine the level of risk posed by the cutoff date of the policy. 

DeFoe began his analysis by gathering raw data for locomotive age and part returns for the CCBII product line. In his spreadsheet model, he applied Crystal Ball’s distribution fitting feature to define probability distributions for the range of expected days to failure of an electro-pneumatic portion and the range of ages of the locomotives from which the parts came. Prior to this, NYAB estimated a single value based on intuition and experience to represent these uncertain variables. With these new probability distributions, DeFoe could better address the unknown relationship between the cutoff criteria for the rebuilding and placement of the mechanical part and the ages of locomotives that required a replacement part. 

He set the baseline using the 120- day cutoff for rebuilding the portion and simulated the Excel model for 10,000 trials. His forecast (output) cell was simply a formula that pulled a random locomotive from the failed locomotives distribution (age) and subtracted a random part from the failed part distribution (age). Forecast results with a negative age indicated that a mechanical part with some number of days of use had been placed on a newer locomotive (one with less days of use). The simulation showed that this scenario occurred for about 1.6% of the mechanical units.

Because the company’s target was to have less than 5% older replacement parts on younger locomotives, DeFoe determined that NYAB could extend the window for rebuilding parts to 210 days. The analysis also showed that 2.25% of the returned electropneumatic portions came in before 120 days and 7.2% came in before 210 days. This meant that NYAB could triple the number of units exempt from rebuild and still meet their goal of <5% of unit age exceeding locomotive age. The simulation results led to a threefold reduction in pneumatic rebuilds and an accompanying cost savings. But even beyond that, we now know the risk up front, which no one knew before, said DeFoe. Once the data was input into the spreadsheet, the analysis with Crystal Ball became almost simple.

In addition to the ability to simulate a spreadsheet model, DeFoe appreciates the Crystal Ball report-generating feature, which lets him summarize and send all the information to management with little to no work. The manager I did the CCBII analysis for was very pleased with the overall analysis, timeliness, output, and flexibility. The rest of management was glad to see a circumstance where they had to expend no money to gain a sizeable cost savings. Two other features DeFoe finds helpful are the fitting of the distributions to real data and the graphical interface, which simplifies how he sets and adjusts cutoff ranges.

Another application for Crystal Ball at NYAB is simulating part tolerances. When an engineer needs to introduce a design change, either due to a problem part or for a cost reduction, he first builds and then tests a number of assemblies with parts whose characteristics (e.g., spring rates, diameters, depths, squareness) are known. The test values and measurements are fit to Crystal Ball assumptions and simulated. Using the results and the desired test range, the part tolerances are adjusted to meet the target. We may not be die-hard users of Crystal Ball, but I believe that when we use it, the benefits are great, said DeFoe. If nothing else, we can walk into a change knowing our risk and probability for success.