I’ve spent many years troubleshooting equipment, along with investigative research. In research-statistical design, data analysis is frequently used when test information is available. This article shows how very basic data analysis can be quite useful in troubleshooting.

Most specialists are knowledgeable about plotting histograms and the value obtainable from such graphs as represented by the distribution of collected data. Remember: “When looking for an answer, a good place to start is to follow the data.” In this case history, we will review a production line that machines a precision groove in a casting bore for automotive use. Hundreds of thousands of casting bores are machined annually, and the groove out of roundness must be maintained within ± 0.003 in. on the diameter since a seal fits within the groove.

Metrics matter

Fig. 1 shows one of the many clamping fixtures on the production line, which holds the casting for machining. To analyze clamping forces for this application, a load cell was designed to replace the part in the assembly and measure the clamping force. The load cell was moved from fixture to fixture. Periodically, a casting would be machined “out of round.” It was believed that the fixtures might be clamping too tightly or too loosely, thus causing the rejects.

Fig. 1. Clamping fixture assembly.

The fixture is designed so that, when torque is applied via a hydraulic motor, as shown in Fig. 1, the part is clamped into place by screw action. A force, F, clamps the part securely in place. With too much clamping force, the part can distort and squeezed into an oval shape. When the grooving tool makes the groove, it is a true circle. When the part is “squeezed” too much, it becomes distorted when unclamped. Result: The part is rejected.

A known torque of 300 in.-lb was applied to each fixture, and the force measured on 215 fixtures. Fig. 2 shows the distribution of the measured parts’ quality from this study. A statistical analysis of the data determines the interaction of the many variables; however, one observation is obvious from Fig. 2—the clamping force varies too much from fixture to fixture. Since the force is directly related to the out of round of the groove, this condition directly contributed to the reject rate. The machining operation is designed for a 2,000-lb to 3,000-lb clamping force.

Fig. 2. Clamping force distribution results on   manufacturing assembly.

Root cause and solution

Other causes contributed to the rejection rate and poor quality of the parts. But the worn-out fixtures and poor calibration of the torque motors were the leading factors of poor quality and low production. Making those needed repairs reduced the reject rate by 70%.

Obviously, these manufacturing operations are not directly applicable to the hydrocarbon processing industry. However, specialists should understand that even basic plotting of data can provide valuable troubleshooting information. In this case, spread of the data represents a deviation from the norm. The troubleshooting process should then question what has caused such a spread.

Next month

In “Case 72: Statistical visual data can be useful in troubleshooting—Part 2,” the example investigates how a simple analytical model could have been applied before any data is collected. This example illustrates the effect of friction and torque on the clamping force. HP