One day my boss tells me that one of our inverters, a 5 kW DC to three-phase AC unit, has too many service calls and asks me to see what I can do to lower the failure rate.
The inverter in question is used to power the air brake compressor. It is approximately the size of two big microwave ovens stuck together. The air compressor is essential to the proper operation of subway cars that use an air brake system . In this system air pressure is used to control the brakes on all the cars in a train. When the compressor inverter fails, the compressor cannot pump air into the brake air reservoir and the whole train has to be taken out of service for repairs.
The inverters were returned to our factory for repairs. One failure mode stood out and it accounted for 16 % of all repairs. The part that failed the most is a small signal transistor—part of the inverter fault alarm circuit. That circuit drives a train line and lights up the air compressor fault lamp on the driver’s status panel.
A train line is a line that goes to all the cars in one train. There are control train lines and alarm train lines. The fault train line lights up the incandescent lamps (the car design is from the mid-1990s) in the driver cab of all the cars in a train. In the train six lamps in total are connected to that line.
I went back to the shop floor and asked the technicians for the failed parts. They told me that the failed parts are discarded after repairs. All I had access to were the schematics.
I analyzed the circuit. An isolated open collector signal drives the base of Q1, with a 10 kilohm pull-up resistor, R1, connected to the battery voltage. Transistor Q1 is a NPN Darlington similar to the TIP122. The Darlington collector, Q1C, is connected to the battery supply, 74 V, and its emitter, Q1E, goes to a current sensing resistor, R21, 700 milliohms. Finally a diode, D1, connects to the train line. The base of the current limiting transistor, Q2B, is connected to the Q1E-R2 node and Q2’s emitter to R2-D1 anode. Transistor Q2’s collector is tied to Q1’s base. This is a standard current limiter circuit  . The transistor that failed the most was Q2 and it was similar to the MPSA42.
Incandescent lamps have significant turn-on surge current. The ratio between hot and cold filament resistance varies between 12:1 and 18:1 . This current limiting circuit protects transistor Q1 against short circuits and lamp turn-on current overloads.
The question is: what was causing Q2 to fail?
Could it be because the collector to emitter voltage was too high? No, Q2’s collector was connected to the base of Q1 and Q1’s emitter was connected to the bottom of R2. The maximum Vce is three times Vbe, or about 2.1 V. Transistor Q2’s maximum Vceo is 300 V. The same reasoning applies to Q2 Vcbo with only two Vbe drops.
Could Q2 fail because of too much collector current? No, the pull-up resistor combined with the battery voltage can only supply a maximum of 7.4 mA and Q2’s maximum continuous collector current is 500 mA.
Could Q2 fail because of too much power dissipation? No, with a maximum collector-to-emitter voltage of 2.1 V and a maximum current of 7.4 mA, peak power dissipation is less than 20 mW and Q2’s maximum rating is 625 mW.
Could Q2 fail because of reverse current or voltage? No, diode D1 protects against that.
The only cause of failure left was excess base current into Q2. The maximum allowed base current is rarely given in manufacturer’s specifications. One source I found is for the 2N3904—a transistor in TO-92 package—and the maximum value for peak base current is 100 mA .
How could the base current into Q2 be so high in this circuit? Transistors do not switch state instantly. With Q1 turned on and saturated, if a short circuit occurs it takes time for Q2 to turn on and for Q1 to turn off. Apparently, these delays made it possible for the base current in Q2 to be higher than 100 mA and cause permanent damage. Until Q2 has turned on and Q1 turned off, nothing is limiting the current in the circuit except the circuit resistance made up of the 0.7 ohm resistor. Adding to the problem is the exponential relationship between the base-emitter voltage and the base current. A 60 mV increase in base-emitter voltage in Q2 causes a 10-fold increase in its base current .
How could I limit the base current into Q2? The solution is simple: insert a small valued resistor in series with Q2’s base.
Will this change the current limit setting? Calculations showed that a 220 ohm resistor in series with Q2’s base produced an acceptable 6% increase in the limiting current value. This resistor will also protect Q2 when Q1 fails short-circuit collector to emitter.
To simplify the repairs I selected a transistor with a built-in base resistor with the same pin-out as the original. The only changes were to the parts list and the schematic; no modifications were needed on the PCB.
I informed my boss that I would require that all units that came into the shop be retrofitted. He agreed with me. I issued a request that when a unit was returned for service, transistor Q2 was to be changed to the new part whether Q2 was functional or not. After that, not a single modified unit returned with a failed Q2.
The lessons learned are: 1) beware of unlimited current getting into a BJT base terminal; 2) it is not because there are no limits given in the data sheet to a device capability, that the rating is infinite, and; 3) always keep failed parts on hand in case further analysis is needed.
 See the article on Railway air brake on Wikipedia, https://en.wikipedia.org/wiki/Railway_air_brake
 Horowitz, Paul and Winfield Hill. The Art of Electronics. 3rd ed. improved, Cambridge University Press, 2015. Appendix B, figure B3. If you have the second edition see appendix E figure E3.
 Typical current limiter circuit, https://electronics.stackexchange.com/questions/195173/bjt-current-limiter-for-linear-power-supply
 General Electric, SCR Manual Including Triacs and Other Thyristors, fifth edition, 1972, table 9.1, page 245.
 Philips Semiconductor, data sheet 2N3904 NPN switching transistor, Product specification 2004 Oct 11. https://datasheet.octopart.com/2N3904-Philips-datasheet-71790.pdf
 Tektronix, 577-177-D1 or D2 Operators Instruction Manual, document 070-1436-00 dated 1172, page 23, figure 19. This document contains one of the rare figures that show measurements of collector current versus base to emitter voltage over nine decades from 200 pA per division up to 20 mA per division. We can see the slope of 60 mV of Vbe per decade of collector current and thus base current. The caption refers to the transistor as a 2N3509; the text in paragraph above refers to a 2N3905, so there is uncertainty about the exact transistor type used.
Daniel Dufresne is a retired engineer and has worked in telecommunication, mass transit, consumer products, and high power electronic design. He also was a professor at Cégep de Saint-Laurent. He lives in Montreal, Canada and still works on electronic projects.