Take the 2-minute tour ×
Programmers Stack Exchange is a question and answer site for professional programmers interested in conceptual questions about software development. It's 100% free, no registration required.

I am working on a CNC (computer numerical control) project which cuts shapes into metal with help of laser.

Now my problem is once in a while (1-2 times in 20 odd days) the cutting goes wrong or not according to what is set.

But this causes loss so the client is not very happy about it.

I tried to find out the the cause of it by

  1. Including log files
  2. Debugging
  3. Repeating the same environment.

But it wont repeat.

A pause and continue operation will again make it to run smoothly without the bug reappearing.

How do I tackle this issue? Should I state it as a Hardware Problem?

share|improve this question
15  
Welcome to the wonderful world of the heisenbug *8') –  Mark Booth Apr 3 '12 at 10:14
    
When you say it happens 1 to 2 times in 20 days, does this mean that it takes about 20 days for it to appear or it sometimes appears after day 1, sometimes day 3 etc... –  Dunk Apr 3 '12 at 18:57
    
@Dunk there is no specific timing to it, but never appeared in a week twice so far. –  Shirish11 Apr 4 '12 at 4:33
    
@Shirish - I was leaning towards a clock overflow problem not being handled properly which I've seen a couple of times on systems whose problem seems to occur every so many days and upon further inspection, exactly every so many days (or multiple thereof). –  Dunk Apr 10 '12 at 18:52
    
What is happening while the system is paused? What memory/counters/hardware are still changing? What about when you continue? It seems like whatever changes while you do those operations is a clue to the cause of the problem. –  Dunk Apr 10 '12 at 19:01
add comment

7 Answers

up vote 23 down vote accepted

Work arounds

As ChrisF suggests, the pragmatic short term solution may be to use the pause and resume trick, but you have to talk to your customers to know what your priorities should be. For example:

  • If the fault trashes a £1000 part or causes 4 hours of downtime once a week, while the pause-resume fix reduces production by 1%, they will probably prefer the fix right now.

  • If the fault trashes a £1 part or causes 4 minutes of downtime once a week, but the pause-resume fix reduces production by 1%, they will probably prefer to wait for a fix which doesn't affect production rate.

Having worked in the laser micro-machining industry for many years, I know just how much pressure you can be under to optimise the process and make your machine produce as many parts per hour as is possible, so either way you are going to be under pressure to fix the problem properly.

Logging

In my experience, the only way to effectively track down a Heisenbug is copious logging. Log everything in and around the part of the code which could be responsible for the error. Learn how to read your log files effectively, make sure you are monitoring following error on your motors (are your stages moving where they should when they should?). Look at the memory usage on the machine, is a memory leak causing a critical process to be starved?

Make sure you are logging user actions too, are you sure that the operator isn't hitting the emergency stop so they can pop out for a shifty cigarette break while it's being fixed? I've seen this happen!

Static analysis

Also, look for correlations between scribing certain patterns and the bug being triggered more or less often. If you can find patterns that trigger the problem more frequently (or never trigger it) these may point to your problem.

Try to make patterns that trigger the problem even more frequently. If you can find a way to trigger the problem reliably then you are half the way to a solution.

Other options

Finally, don't be quick to blame the hardware, but never assume that it's perfect. Many times I've been blamed for problems which turned out to be electrical or mechanical in nature, so you always have to have that at the back of your mind.

Even though you may not normally have access to the machine, remember that some problems can only be efficiently solved on the machine. Sometimes a few days on-site can be worth weeks via remote desktop and months off-line completely. If you run out of off-line options, don't be afraid to propose a site visit, they can only say no.

You might also want to look at the questions and answers to What do you do with a heisenbug? and What to do with bugs that do not repro? but these might not be so useful for your situation.

share|improve this answer
    
more to add to my problem I dont have the hardware at my disposal. And the client is not that educated to understand these programming terms.So hanging on to his system remotely not possible. BTW thanks for the advice will try a work around. –  Shirish11 Apr 3 '12 at 13:17
add comment

The problem is a real one with real consequences for the user - i.e. ruined work etc. so it needs fixing. However, it doesn't have to be fixed "properly". You state:

A pause and continue operation will again make it to run smoothly with the bug reappearing.

In that case just do this. The customer will be happy that they are not wasting material on defective runs even if normal runs take a couple of seconds longer.

Obviously in the long term you might need to fix this "properly" but for the time being cut your losses, go with the workaround and get onto something else.

share|improve this answer
add comment

Rule 1 number one in debugging: you need a reproducible scenario.

If you don't have one, you should work on that first. Can you reproduce that bug in some kind of "simulation mode" of the machine, where no metal is actually cut? This seem to make sense here. Can you run several different cutting programs quickly and automatically, simulating the process of 20 days in a few minutes? That may increase the probability of the issue showing up.

Then, when you have such a scenario, the next step is to gather as much information as possible and actually start debugging.

share|improve this answer
    
simulating the process of 20 days in a few minutes thats not possible. I have to consider the hardware. –  Shirish11 Apr 3 '12 at 11:58
2  
I've never come across a heisenbug that could be reproduced using a simulation mode. The problems are almost always in the components which are simulated out or the coupling between them. As I said, if you can reliably reproduce the problem, you are half the way to a solution. –  Mark Booth Apr 3 '12 at 14:15
    
@Shirish: "simulating the process in a few minutes" may be one extreme, but waiting 20 days for the bug to occur and cut a lot of metal to let the bug pop up is obviously the other extreme. Perhaps there's something possible in between. –  Doc Brown Apr 3 '12 at 14:43
2  
@shirish-if you haven't abstracted away the hardware so that it becomes possible to simulate it means that the design is lacking. It also means that your system could not have been adequately tested. Thus, it is no surprise that the system has issues. –  Dunk Apr 3 '12 at 18:52
1  
@Dunk - Have you ever worked in the laser scribing industry? You don't always have the luxury of a simulator and even if you had a good one, it wouldn't be cost effective to fully simulate all of the intricacies of a complex mechatronic system. Following error, velocity profiling, pulse tracking all at sub-micron precision, interactions between soft & hard real-time systems, Takt time pressure - simulating that lot in real-time would take a cluster, let alone doing it in 1/10,000 of real-time. Faster/better/cheaper - you can rarely have all three, so please try not to be so judgemental. –  Mark Booth Apr 5 '12 at 21:29
show 4 more comments

I had a bug in a game that happened only 1 time in a billion. Fortunately this meant I was seeing it every 15 to 30 minutes, but stepping through the code in the debuggger was not going to work. I ended up putting in debug messages. They needed to use fancy if-statements because I only wanted something when there was a problem. In most cases the debugging code was repeating calculations in the regular code but using different techniques. The repeats did not have to be precise. If I knew a number should always be under 10,000 and it seemed to hit 150,000 on occasion, I'd just check for a value over 100,000. Each time the bug occurred, I'd study my results, devise more elaborate debugging messages (or more precisely, more elaborate checks to see if I should display a message), and wait for the problem to arise again.

Your cycles are going to be a lot longer than mine were, but you will eventually close in on the problem. I do hope you can find the solution by some other, faster method, but this will catch it eventually if nothing else does, and will give you a sense that you are doing something until you come up with a better idea.

(In case it's helpful, I finally solved my problem by cleaning up the few lines of code I finally identified as the problem. I'll swear there was nothing wrong with them, but I think both the optimizer and the CPU were reordering instructions for performance, and I think once in a while they were taking chances to get a bit of extra speed. Even a single core multi-processes these days, and I think every great once in a a while a register got read before it was written to. I switched all calculations to work with local variables. "Instance field" values were moved to local variables right at the start, and the local values were moved back only at the very end, inside synchronization blocks. And I used a local value for the method return value rather than the "instance field" I had been using.)

share|improve this answer
    
+1 for sanity checking and the iterative improvement of logging messages to converge on the root of the problem. –  Mark Booth Apr 3 '12 at 17:17
add comment

Not sure what language this is run in, but if I experience erratic bugs in my code (C++), I will use a tool like valgrind or cppcheck to ensure nothing is going on memory-wise.

share|improve this answer
add comment

I'm going to make an off-the-wall suggestion.

Go to the factory manager and ask to see the power line monitor records for that tool, or that area, for the times when the malfunctions occurred. Also ask him if there was any welding, or any other unusual activity, at around those times.

Several decades ago, my father was having a hell of a time with a minicomputer that was crashing for no reason at all. They called the manufacturer's customer rep.

The rep came into their office, in the factory area, and plugged a voltmeter into the wall, next to the mini, and then said "Watch this."

A few minutes later, the voltmeter suddenly sagged, significantly, then came back. The rep said "That was him striking his test arc. Wait a minute." Shortly after that, the voltmeter sagged again, and this time it stayed sagged.

The rep said "That's your problem. You've got a guy welding on the factory floor, and he's on the same power leg you are. I saw him setting up as I was walking in."

They had to run a completely separate power feed to the office.

share|improve this answer
add comment

An extension on RalphChapin's answer:

Over the years I have had to hunt a fair number of bugs that only showed themselves on systems I couldn't duplicate because of attached hardware.

In addition to logging like crazy one other thing I found useful: Putting information on the screen showing where the code was and values of some relevant variables. When the problem showed up even the factory floor workers could read me the information.

It usually took a few rounds of refinement to pin it down exactly but it was very effective.

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.