Overcome Hidden Measurement Errors

You often need to become a detective to find and remove unusual errors.

Joe Tauser, Joseph Tauser & Associates, St. Louis, MO -- Test & Measurement World, 12/1/1999

Test engineers usually measure small resistances by connecting a four-wire Kelvin ohmmeter to a sample and reading the display. But we encountered significant errors when we applied this seemingly straightforward technique to measure the resistance of continuously moving wire. Problems arose from a common phenomenon known as the triboelectric effect. We also discovered that engineers can’t always build gear that duplicates the functions of high-quality test equipment.

We needed to measure the resistance of silver plated onto copper wire that moved through an electroplating bath and then past measuring electrodes (Fig. 1). The plating system applied a constant current to the bath and then varied the wire’s speed through the bath to achieve the specified resistance. After going through a final rinse, the wire passed over a V+ and I+ probe, and exactly 2 ft later the wire passed over a V– and I– probe that completed the Kelvin bridge. We divided the resistance reading by 2 to get the correct resistance per foot. Depending on the plating required, the desired resistance of the silver-plated wire typically ranged from 0.050 to 2.000 W/ft. The plated wire had to have a resistance variance of ±0.015 W.

Our first measurement system supplied a 100-mA excitation current—monitored with a signal transducer—to the wire. Another signal transducer monitored the voltage across the 2-ft wire. An RTD provided a signal so we could compensate for temperature in the measurements.

We used Allen-Bradley 1746-NI4 analog-input cards with 14-bit resolution, and we calculated that we had a 0.003-W uncertainty—plenty of room within the ±0.015-W variance we were allowed. We calibrated our system and started to plate wire.

For the first batch of wire we plated, our instruments measured a resistance of 1.000 W/ft, just what we wanted. But the quality lab said the wire was barely within spec; it was consistently on the high side of the 0.015-W tolerance. So what we measured as 1.000 W, the lab measured as about 1.015 W. How could a 0.003-W measurement uncertainty turn into a 0.015-W measurement difference?

The next day, we put calibration resistors on the line before starting production, but we no longer got good readings. The system seemed to lose calibration overnight. We had “fun” recalibrating the system every day before doing our tests only to achieve the same result: The resistance always was higher than the computer average, to varying degrees.

Try a Commercial DMM
Perhaps we had to admit that we couldn’t build our own Kelvin system. We bought an HP 34401A
6 1/2-digit DMM and a Basic module so the Allen-Bradley controller could “talk” to it. Initially, we got encouraging results. The meter read all the resistance standards perfectly, and gathering the data through the Basic module meant we no longer had to calibrate the line daily. We tried plating wire again, this time with a resistance of 100 mW/ft, but the wire’s actual resistance wasn’t what our measurement system said it was.

We noticed that the resistance of the wire fluctuated as soon as the line started, even in the absence of the plating current. We set the DMM to its DC-voltage mode and measured the voltage across the wire. The DMM read 0 V with the wire stationary, and then measured a small voltage that increased to about 100 µV as the wire speed increased. This voltage also varied as the curvature of the unspooling wire varied the surface contact of the wire to the probes. Here was the problem!

Figure 1. A wire-plating system uses a four-wire connection to measure the resistance of plated wire. Based on the measurement value, a PLC controls the wire’s speed to maintain a constant, preset resistance.

Checking the specs for the HP DMM revealed a 1-mA excitation current. One milliamp of current passing through the 100-mW wire produces a voltage of 100 µV, the same amplitude as the mysterious voltage.

Don’t Give Us Static
A friend who has worked with electricity almost since Edison’s time told me matter-of-factly that I was observing the triboelectric effect, a fancy name for static electricity. He told me any time two materials rub together, they generate a small voltage. To overcome the tribolelectric effect, the excitation signal had to be much larger. Unfortunately, the HP DMM was limited internally to producing a 1-mA current, so we had to choose another instrument.

We borrowed a Tegam Model 1750 resistance-measuring system. This unit offered two helpful features. First, it let us program the bias current, and it could supply up to 1 A for the 200-mW range.

Second, it provided bipolar excitation. That is, the unit reverses current flow several times a second and averages the resulting voltage measurements to calculate a resistance. We quickly reprogrammed the Basic module to communicate with the Tegam meter.

Because the Tegam meter effectively produced an AC signal, we had to consider the characteristic impedance of the 50-ft wires connecting the meter and the plated-wire test contacts. We used an LCR meter to compare the high-quality leads we got with the meter, made of high-quality microphone wire, to our existing 16-gauge shielded twisted-pair wires and found a significant capacitance/ft difference. We changed our connections to use the audio wire and settled on a 100-mA excitation current. Then, we verified the proper operation of the system once again with our calibration resistors.

We set up the measurement system for 100-mW wire and started the plating line, quite pleased that the reading didn’t jump when the wire started moving. We took the finished roll to the quality lab, and it was right on spec. T&MW

Joe Tauser is a licensed professional engineer who graduated from the University of Missouri-Rolla with a B.S.E.E. He works as a system integrator specializing in automatic test systems using PLCs and PC-based data-acquisition equipment.