LM358 failed after 30 minutes and gets hot in 4-20mA pump circuit

[djzulu]

I have a current pump circuit that converts 0-5V to 4-20mA.  When the circuit was online for 30 minutes the LM358 got hot and finally blew. I thought that the cause could be the possible high temperature environment and replace the LM358 with the high temperature variant LM2904.

After changed the opamp to the LM2904 I found that after 30 mins of operation the LM2904 would blow and get hot. There are some high voltages in the proximity and I wanted to add TVS protection, it seems that with TVS if the voltage is below the threshold adding the TVS would have no affect unless a TVS event occurs.

Would the attached circuit cause the LM2904/LM358 to fail without a transient event ?

What is the best way to protect the circuit with TVS diodes ?

P0-0 = +24V DC

" alt="" class="bbc_img" />

In addition I have included a schematic showing the TVS devices.

[srb1954]

You stand a better chance of getting an answer if you post a better quality schematic that is sufficiently readable so we can actually see the component values and other details.

[djzulu]

I've updated the Schematic.

[bdunham7]

What package are you using?

[djzulu]

All components are SMC, the LM2904 is SOIC-8.

[bdunham7]

At first glance it doesn't look like you're actually over thermal limits unless your ambient is pretty high, like 50-60C--then you'd be getting close.  However, the way it fails does seem like a thermal issue.  Reducing the supply voltage would help if that is feasible with your design.  24V is pretty close to the design limit as well.  What compliance voltage is required for the 4-20mA output?

[djzulu]

The initial PCB had the LM358 op-amps, the card was in a terminal box that could have experienced temperatures up to 60C in summer. After a few failures we realized that the thermal limit was too low and opted to use the high temperature version of the LM358, the LM2904 which is rated to 125C. After replacement of the boards in winter when the temperatures were close to freezing we experienced failures. The failure mode was interesting, the boards would operate fine, some would run for weeks and others would fail almost immediately, once failed the LM2904 would get too hot to touch.

There are some coils in the vicinity that have secondary voltages in the 30-50kV range, to my best bet there is a transient event that is blowing the LM2904/LM358 and causing an internal short which is raising the temperature.

Note we have other op-amps on the board, i.e MCP6002 that have never failed, these appear to be just as exposed to transients are the LM358's are.

The 4-20mA voltage lop does not have a required voltage, anything between 12 and 24V would work, we are measuring a voltage over a 475ohm shunt resistor.

[bdunham7]

You might just try putting a ~6V zener in series with your supply to drop the voltage down to 18V.  If your failures stop, that tells you where the problem is, if they continue, then you know where it isn't.

[Swainster]

When specifying TVS diodes, you need to allow some headroom for component variation - perhaps something like 10% i.e. for the 5V regulator, your Vrwm should ideally be above 5.5V, which would bump your TVS diode up to the next available size. That said:

1) You seem to straying quite close to the datasheet limits of lm358/2904, and as such should be careful with part substitution - not all manufacturers' lm358/2904s will have exactly the same limits.
2) Seems like this circuit is subject to proper industrial conditions, and mounted on the end on a cable. This adds a whole extra dimension to the design requirements. Hard to know where to start...
3) Current loop transmitters are typically supplied from the far end of the cable (receiver end) - is this the case here?

[djzulu]

I am not sure why I am at the limits of the LM358, it does sound plausible because only the lm358's are failing. The maximum supply valve voltage from the datasheet is 32 volts and I am supplying the LM358 from the board with 24 volts.

The supplier is different to loop powered 4-20 milliamp circuits, we are powering the op-amps at the board. The signal cables are typically 400 m long and I'll terminated with a shunt resistor to ground. The shunt resistors value is 360 ohms (475R was error), this provides a measuring range of 1.44V to 7.2V.

The input BUF is from an op-amps power by 5V.

[magic]

Simply driving 20mA DC output causes the chip to dissipate 17V·20mA = 340mW.

Are you sure it isn't oscillating at a few hundred kHz on top of that?

[djzulu]

The lm358 is that 4 milliamps for the most part, the 20 milliamp output is a seldom event and it's a peak output when it occurs.

I have not put an oscilloscope on the output but we will try that.

[Andy Watson]

The signal cables are typically 400 m long and I'll terminated with a shunt resistor to ground.

Are these complete loops or is the remote ground connected to a common point that is assumed to be 0V?

[djzulu]

The remote connection is connected to ground 0v at the shunt resistor. We monitor the voltage over the shunt resistor using an ADC.

[djzulu]

You might just try putting a ~6V zener in series with your supply to drop the voltage down to 18V.  If your failures stop, that tells you where the problem is, if they continue, then you know where it isn't.

I do not understand how a Zener in series would reduce the voltage, I can reduce the voltage at the power supply from 24V to 18V and try to fail the components.

[Kean]

The remote connection is connected to ground 0v at the shunt resistor. We monitor the voltage over the shunt resistor using an ADC.

Are you sure there isn't any potential difference (e.g large transients) between the "grounds" at each end of that 400m cable?
Do you see the same temperature rise and/or failures when the output is just terminated with a shunt resistor and no remote measuring equipment?  (with or without the long cable)

[Kean]

You might just try putting a ~6V zener in series with your supply to drop the voltage down to 18V.  If your failures stop, that tells you where the problem is, if they continue, then you know where it isn't.

I do not understand how a Zener in series would reduce the voltage, I can reduce the voltage at the power supply from 24V to 18V and try to fail the components.

A reverse biased zener in series between P0-0 and the opamp power supply pin will provide a voltage drop.  You'd want to have a local decoupling capacitor for the opamp (you don't show any in the schematic).

[bdunham7]

I do not understand how a Zener in series would reduce the voltage, I can reduce the voltage at the power supply from 24V to 18V and try to fail the components.

OK, if you can just reduce the power supply voltage some other way for testing, that's good enough.

Yikes, a 400m cable?  What form exactly is the cable--a twisted pair?  What kind of environment is it in? Do the failures occur in testing without that cable?  Can you connect an oscilloscope and record what goes on at the output of your current source? 

[free_electron]

are you sure they are not oscillating at very high frequency ( MHz ). where is your power supply decoupling ?
Opamps can be unreliable when the input difference gets large. put two diodes antiparallel between the inputs. that way there can never be more than 0.6 volts of delta.

[wraper]

The initial PCB had the LM358 op-amps, the card was in a terminal box that could have experienced temperatures up to 60C in summer. After a few failures we realized that the thermal limit was too low and opted to use the high temperature version of the LM358, the LM2904 which is rated to 125C.

That was quite pointless, it's exactly the same IC but with tighter binning. It's just a guarantee that IC will meet specs in wider temperature range. Maximum junction and storage temperatures are exactly the same for all variants.

[djzulu]

The remote connection is connected to ground 0v at the shunt resistor. We monitor the voltage over the shunt resistor using an ADC.

Are you sure there isn't any potential difference (e.g large transients) between the "grounds" at each end of that 400m cable?
Do you see the same temperature rise and/or failures when the output is just terminated with a shunt resistor and no remote measuring equipment?  (with or without the long cable)

Under steady state conditions there is no temperature rise, the temperature seems to spike just before a failure.

[djzulu]

are you sure they are not oscillating at very high frequency ( MHz ). where is your power supply decoupling ?
Opamps can be unreliable when the input difference gets large. put two diodes antiparallel between the inputs. that way there can never be more than 0.6 volts of delta.

We will add the diodes and a decoupling capacitor to the +24V rail, as the failure occurs with a lab power supply we do not believe that this is the issue, we do not know the reason for the failure.

[Swainster]

Being able to duplicate the fault on the bench is very interesting - it would take most of the environmental conditions out of the equation. Presumably this was with the receiver closely connected to the transmitter?

Talking of the receiver, is this a standard loop receiver or did you roll your own? If it is an off the shelf part, is it supplying the loop power? If this is the case then supplying power to the output of an lm358 will quickly burn it out.

Ignoring the bench failure for the moment, with a long enough cable, especially in an industrial environment, if the supplies at either end are not fully floating with respect to each other (e.g. if they are partially or fully tied to local ground), then significant common current will flow in the loop conductors. This could burn out the transmitter (and has been known to melt wires, at least in communication cables).

[djzulu]

Talking of the receiver, is this a standard loop receiver or did you roll your own? If it is an off the shelf part, is it supplying the loop power? If this is the case then supplying power to the output of an lm358 will quickly burn it out.

The received is a simple shunt resistor 360R, we measure voltage of 1.9 to 7.2V over the resistor. I have one card that has been running for 2 months with a shunt on the output in the lab.

Ignoring the bench failure for the moment, with a long enough cable, especially in an industrial environment, if the supplies at either end are not fully floating with respect to each other (e.g. if they are partially or fully tied to local ground), then significant common current will flow in the loop conductors. This could burn out the transmitter (and has been known to melt wires, at least in communication cables).

If I short the outputs this would drive 20mA through the LM358, would this rule out cabling or are you concerned about transient conditions too ?

[Swainster]

Shorting the lm358 will rule out current being driven into the output, though it will heat up the op-amp a fair bit - not all lm358s are rated for constant short at 24V (are any of them?). Typically constant short to gnd for lm358 assumes room temperature and an equal split supply I.e. only shorting about 15V to ground. I guess you would already be pushing abs max power for the package at 15V, never mind at 24V

[djzulu]

Shorting the lm358 will rule out current being driven into the output, though it will heat up the op-amp a fair bit - not all lm358s are rated for constant short at 24V (are any of them?). Typically constant short to gnd for lm358 assumes room temperature and an equal split supply I.e. only shorting about 15V to ground. I guess you would already be pushing abs max power for the package at 15V, never mind at 24V

The Circuit has resistors to allow for a short, we plan to increase the shunt resistors load.

Is there anything else to consider? We still feel that transient voltages from the local HV 45kV coil could be leaking through to the circuit, this we are trying to simulate in the lab.

[floobydust]

Have you run this with a capacitive load? Real world 4-20mA has a lot of cable capacitance, between the two wires as well as the shield.
I always use an emitter follower rather than a lone op-amp to drive 4-20mA but use higher compliance voltage of 24VDC rail. It's generally a catastrophe if the output shorts and goes full tilt, in most plant equipment. So I monitor the ouput voltage well to detect that fault.

[bdunham7]

the temperature seems to spike just before a failure.

That might support the idea mentioned earlier by others that the circuit may be breaking up into oscillations.  You'd need to get a scope hooked up and see.  If it is, you might try increasing C1603, perhaps by quite a lot.

[magic]

Do you see the weird phase dip in the closed loop response of the current sense amplifier (U1601B) near 10kHz? That's C1603
Also Cload effects are involved a little, due to the voltage follower path to "amp" node through the noninverting input of U1601B with C1603 providing unity feedback starting from a few kHz.

I don't get the point of C1603 at all.
That being said, none of it obviously looks like the whole circuit should oscillate just yet. I think.

On the other hand, C1602 could be increased.

[djzulu]

the temperature seems to spike just before a failure.

That might support the idea mentioned earlier by others that the circuit may be breaking up into oscillations.  You'd need to get a scope hooked up and see.  If it is, you might try increasing C1603, perhaps by quite a lot.

If I reduce C1603 will that increase the oscillations and help to validate the failure mode?

[djzulu]

Do you see the weird phase dip in the closed loop response of the current sense amplifier (U1601B) near 10kHz? That's C1603

I don't get the point of C1603 at all.

On the other hand, C1602 could be increased.

For our application we are really not interested in any of the frequencies that are above 1 kilohertz, there is actually a smoothing capacitor on the 5 volt input to the circuit The smooths the input to about 5 Hertz. The reason is that this is a very simple circuit that takes the output of an instrument in voltage 0 to 5V and converts it to a 4 to 20 milliamp signal for transmission.

I believe that the original intention of the capacitors was to spoothen out the output if this is causing harmonics then we need to change them I can certainly increase the value of c1602 and remove c1603.

Does this explanation help regarding the oscillations and is there a way to clamp these down so that there is no possibility of oscillations causing the LM358 failure?

[magic]

But can you see what C1603 is doing?

It decreases closed loop differential gain of U1601B, which means that U1601C is mislead to believe that transmitted AC current is lower than in reality. If you remove the filter and test with 10kHz signal, you would likely find that the regulator transmits more current than at 10Hz, for the same input amplitude.

Furthermore, it ruins AC common mode rejection. Common mode signal present at the positive end of R1613 is simply passed to the output of U1601B at unity gain, due to direct feedback through C1603. This causes all sorts of extra oddities, dependent on load reactance.

The whole thing looks like its exact behavior is needlessly difficult to predict. But no, I don't have any sim example where it clearly oscillates. The phase dip I have shown is not -90°.


I don't think there would be much wrong with:
- removing C1603, to make U1601B work as proper differential amplifier, giving accurate (amplified) image of R1613 current, up to tens of kHz that this chip ought to be capable of
- increasing C1602 to a few nF (or more) to dumb down U1601C a little more; above a few kHz it will simply be a voltage follower, ignoring any feedback coming from U1601B

[jwet]

Have you heard of Action Paks?   They are little isolated voltage converters that do conversions like this- zero to 5 to 4-20 mA is a standard function.  They are completely isolated with their own internal supplies for the low voltage input and the output. They come in old style relay packs and with a socket on the bottom that can go into a standard relay socket for screw terminations.  They're calibratable with a span and zero pot. This is the kind of thing you need- with Isolation.  There are other vendors and they're available for DIN rails too.  I think this little pump with a bare op-amp output heading out into a dirty plant driving a bunch of cable needs more than a tweak of some caps values.  We haven't talked about grounds yet, I suspect this is the real issue.  If you take your delicate little op-amp output, go through a bunch of cable and then a single ended sense resistor that is grounded at the receiver, you just negated most of purpose of 4-20 mA.  4-20 mA loops need to exist as loops with the sender floating and the return current coming back to the sender.  You can establish a ground at only one point.  This application requires isolation.  You also have your input referenced to ground which is getting pulled or pushed by the remote receiver's ground.  Grounds in a industrial environment 400m away have nothing to do with  your ground.  I suspect that there are ground imbalances that are pulling that little LM358 way out of its abs maxes.  Put in an Action Pack and call it a day- fully isolated problem solvers.  Keep this circuit as a summer intern project for the future and maybe a lesson on 4-20 mA loops that the student will never forget.

You could use a 4-20 mA to 4-20 mA to preserve this circuit on your board.  Run your 4-20 mA in and let it produce isolated 4-20 mA for the industrial run.  Cleaner probably.

[floobydust]

It looks like OP using a PWM DAC with a 1st order LPF. What carrier frequency? I would expect a lot of HF ripple coming in to the op-amp which might aggravate things.

[djzulu]

Have you heard of Action Paks?   They are little isolated voltage converters that do conversions like this- zero to 5 to 4-20 mA is a standard function.  They are completely isolated with their own internal supplies for the low voltage input and the output. They come in old style relay packs and with a socket on the bottom that can go into a standard relay socket for screw terminations.  They're calibratable with a span and zero pot. This is the kind of thing you need- with Isolation.  There are other vendors and they're available for DIN rails too.  I think this little pump with a bare op-amp output heading out into a dirty plant driving a bunch of cable needs more than a tweak of some caps values.  We haven't talked about grounds yet, I suspect this is the real issue.  If you take your delicate little op-amp output, go through a bunch of cable and then a single ended sense resistor that is grounded at the receiver, you just negated most of purpose of 4-20 mA.  4-20 mA loops need to exist as loops with the sender floating and the return current coming back to the sender.  You can establish a ground at only one point.  This application requires isolation.  You also have your input referenced to ground which is getting pulled or pushed by the remote receiver's ground.  Grounds in a industrial environment 400m away have nothing to do with  your ground.  I suspect that there are ground imbalances that are pulling that little LM358 way out of its abs maxes.  Put in an Action Pack and call it a day- fully isolated problem solvers.  Keep this circuit as a summer intern project for the future and maybe a lesson on 4-20 mA loops that the student will never forget.

You could use a 4-20 mA to 4-20 mA to preserve this circuit on your board.  Run your 4-20 mA in and let it produce isolated 4-20 mA for the industrial run.  Cleaner probably.

We know of IC's that perform conversion, these even have regulators built in for powering the circuit. Due to the chip shortage out PCB assembly folks cant get these and we are doing it old school. In reality I don't see why using op-amps would not work, we have this exact circuit working on a slightly different application for 3 years without any issues.

[djzulu]

It looks like OP using a PWM DAC with a 1st order LPF. What carrier frequency? I would expect a lot of HF ripple coming in to the op-amp which might aggravate things.

We are not interested in harmonics, only interested in having a stable constant current circuit for converting the 0-5V board signal to 4-20mA. If we are introducing harmonics it is not deliberate and should be conditioned.

[floobydust]

What is the source of the analog current command? Nice steady pure DC? Circuit diagram says otherwise. It's not harmonics, it's attenuation of the PWM carrier with the LPF.

[djzulu]

What is the source of the analog current command? Nice steady pure DC? Circuit diagram says otherwise. It's not harmonics, it's attenuation of the PWM carrier with the LPF.

The 5V is the output of an amplifier, this amplifies a signal from a Piezo sensor. It is usually stable and changes very little with time.

[djzulu]

Have you run this with a capacitive load? Real world 4-20mA has a lot of cable capacitance, between the two wires as well as the shield.
I always use an emitter follower rather than a lone op-amp to drive 4-20mA but use higher compliance voltage of 24VDC rail. It's generally a catastrophe if the output shorts and goes full tilt, in most plant equipment. So I monitor the ouput voltage well to detect that fault.

I thought that this diagram may help to illustrate the local high voltage that we had thought was leading to a transient voltage spike.

[floobydust]

Assuming the LM2904 is oscillating (or slewing), you can upsize C1601 and see if that helps, try a different (better) op-amp, where is a decoupling cap for pin 8 V_CC?
I would make sure the current command is smooth DC, not HF with noise as the active rectifier is not band-limited.

Your grounding could be a problem as well, the 'common' from the HV return is extremely noisy and full of RF. The 4-20mA cable can act as a RF return as well.
It looks like a car ignition coil, which is going to kill electronics just for fun.
How your PC board is (earth) grounded is critical. I would also check the HV wiring is not leaking/corona.

[vk6zgo]

Simply driving 20mA DC output causes the chip to dissipate 17V·20mA = 340mW.

Are you sure it isn't oscillating at a few hundred kHz on top of that?

358s are rather fond of oscillating.

We had a radio transmitter design, where a LM358 was part of the control circuit to select various power settings.
Several power supplies failed catastrophically while the thing was set to its lowest power setting.

This was counter-intuitive, but we found that on that setting, the 358 was oscillating.
The resulting 78kHz rectangular waveform went "rail to rail" Amplitude Modulating the Tx PAs between cutoff & saturation, and
killing the PSU.
The average power meter was quite happy, however!

A 220 ohm resistor between the output pin & the controlled circuit stopped the oscillation---shame the "rent a EE's" the manufacturer used didn't include that as a precaution.

[djzulu]

Assuming the LM2904 is oscillating (or slewing), you can upsize C1601 and see if that helps, try a different (better) op-amp, where is a decoupling cap for pin 8 V_CC?
I would make sure the current command is smooth DC, not HF with noise as the active rectifier is not band-limited.

The 5V on the pre amp is well decoupled, the 24V VCC on pin 8 does not have decoupling. We are going to add decoupling to the PCB.

Your grounding could be a problem as well, the 'common' from the HV return is extremely noisy and full of RF. The 4-20mA cable can act as a RF return as well.
It looks like a car ignition coil, which is going to kill electronics just for fun.
How your PC board is (earth) grounded is critical. I would also check the HV wiring is not leaking/corona.

The Coil is an ignition coil with an internal switching circuit, I does create a lot of corona, we have been trying to make the coil fail the LM348 to no avail, I have even sparked to the ground pin of the LM348 and it has not failed. That being said the components all need TVS diodes for protection.

[djzulu]

Simply driving 20mA DC output causes the chip to dissipate 17V·20mA = 340mW.

Are you sure it isn't oscillating at a few hundred kHz on top of that?

358s are rather fond of oscillating.

We had a radio transmitter design, where a LM358 was part of the control circuit to select various power settings.
Several power supplies failed catastrophically while the thing was set to its lowest power setting.

This was counter-intuitive, but we found that on that setting, the 358 was oscillating.
The resulting 78kHz rectangular waveform went "rail to rail" Amplitude Modulating the Tx PAs between cutoff & saturation, and
killing the PSU.
The average power meter was quite happy, however!

A 220 ohm resistor between the output pin & the controlled circuit stopped the oscillation---shame the "rent a EE's" the manufacturer used didn't include that as a precaution.

I will do my best to change C1602 and C1603 to eliminate any change of oscillation. My Experience with oscillations is that its not a matter of throwing components at the problem, but rather simulating changes and then testing the board.

Any idea what I should use for C1602 and C1603 instead of the 470pf capacitors?

[djzulu]

But can you see what C1603 is doing?

It decreases closed loop differential gain of U1601B, which means that U1601C is mislead to believe that transmitted AC current is lower than in reality. If you remove the filter and test with 10kHz signal, you would likely find that the regulator transmits more current than at 10Hz, for the same input amplitude.

Furthermore, it ruins AC common mode rejection. Common mode signal present at the positive end of R1613 is simply passed to the output of U1601B at unity gain, due to direct feedback through C1603. This causes all sorts of extra oddities, dependent on load reactance.

The whole thing looks like its exact behavior is needlessly difficult to predict. But no, I don't have any sim example where it clearly oscillates. The phase dip I have shown is not -90°.


I don't think there would be much wrong with:
- removing C1603, to make U1601B work as proper differential amplifier, giving accurate (amplified) image of R1613 current, up to tens of kHz that this chip ought to be capable of
- increasing C1602 to a few nF (or more) to dumb down U1601C a little more; above a few kHz it will simply be a voltage follower, ignoring any feedback coming from U1601B

I ran your LTSpice model and noticed the Oscillation, I tried to change the values of C1602 and C1603 and could not  smooth out the ripple.

[djzulu]

358s are rather fond of oscillating.

Should I remove the two 470pf capacitors ?

[RogerThat]

I would suggest the following.

Groupe R1604 and R1606 in to one 1.2M...reduce BOM.
Remove: D1705
Move: D1706 to be in parallel to C1601. Most likely you can remove this one also, not needed due the high input resistance.
Add( isolation) resistors on both op amp outputs, 25-1000ohm, to avoid oscillations. some op amp don't like to directly drive capacitors. If you don't wont to add isolation resistors you could remove c1602,c1612 for testing.

A trick to greatly improve ESD tolerance is to put small resistor on all inputs/outputs to your chips. Resistors in the size of 10-100ohm helps a lot. In your circuit all inputs/outputs all ready have resistors in front of them except the powersupply...I would investigate the power supply further.

It's not clear from your schematics(where does +MA go? I hope it doesn't go, thru a uninterrupted cable, 400m to the receiver) but make sure you have isolation between the circuit and the line driver...

[RogerThat]

The SMBJ24A have a maximum clamping voltage of 38.9V, I would choose a SMBJ18A instead and adjust supply voltage accordingly. Maximum voltage for the lm358 is 32V. LM2904 is 26V maximum.

[djzulu]

It's not clear from your schematics(where does +MA go? I hope it doesn't go, thru a uninterrupted cable, 400m to the receiver) but make sure you have isolation between the circuit and the line driver...

The 4-20mA is connected to a ~400 Foot (can be meters) cable to a shunt resistor where we monitor the volt drop over a resistor. The cable is powered by the op-amp.

How would I add Isolation between the circuit and the driver?

[djzulu]

Maximum voltage for the lm358 is 32V. LM2904 is 26V maximum.

We are using the onsemi LM2904, this has a max voltage of 32V.

https://datasheet.lcsc.com/lcsc/1811012110_onsemi-LM2904DR2G_C18229.pdf

[jwet]

Look back to post 32- this is what I was talking about.  This is your problem.

[djzulu]

I did not find any IC 4-20mA to 4-20mA Isolation modules to add the the board.

[RogerThat]

You can build an isolator from two optocouplers, an op amp and some resistors. One optocoupler to adjust the current in the loop and one for feedback, the op amp as "controller". There is no need for a specific IC.

Edit: I normal 4-20mA circuits you define who is powering the current loop: sender or reciver. The one who is not powering the loop need to interact with the loop thru an isolator. The isolator prevents current flowing from sender to reciver. You migh have the case where the ground might be hundreds of volts different between the two...you can figure out what happens if that is the case and you don't have isolation.

[djzulu]

Here is a timeline of the failure,



The one channel failed relatively fast, it failed concurrently with a high voltage event.

It failed and then recovered before failing permanently.

We have now added TVS diodes to the inputs and the outputs of the board and will evaluate the board in January.