iSixSigma

Process control with several cavities

Six Sigma – iSixSigma Forums Old Forums General Process control with several cavities

Viewing 25 posts - 1 through 25 (of 25 total)
  • Author
    Posts
  • #35292

    Gabriel
    Participant

    Hi to all.
    I have just requested a supplier to do a thing, the problem is that he asked me “How do I do that?” and I am not sure.
    The supplier injects some rubber parts for us. I asked him “to keep under control the cavity-to-cavity, part-to-part, and long term variations, and to verify that those variations are compatible with the specified requirements”.
    The problems are:
    – The part has 20+ characteristics, many of them crityical to quality.
    – Not allways the most important characterisitics for the application are the more representatives of the “health” of the process.
    – The mould has 24 cavities.
    – The measurements are hard to do and to repeat.
    My crazy ideas (like keeping SPC of several cahracterisitcs in all moulds) are impossible to realize in real-life. It is already hard enough to measure only once each characterisitc of each cavity for mould approval.
    Some brillant ideas?

    0
    #98880

    Anonymous
    Guest

    Gabriel,
    Welcome to the multivariate world … multivariate not just in the sense of multiple characteristics, but also multi-normal of the same characteristic.
    One of the assumptions that many people make is that just because parts are processed together, either in the same bath, the same tool, or in the same web, that the samples are somehow homogeneous process. In my experience this is hardly ever the case. Of course, components of variation assume that parts are random and independent, and normal; although many would dispute that latter restriction!
    My approach has been to follow Taguchi’s lead and consider the process mean and the process uniformity (nominal-is-best.) If you do this, you ought to be able to find out what factors can be used to adjust the shrinkage, and what factors can be used control uniformity.
    This generally works well and I’ve applied this method successfully to many ‘batch’ semiconductor process, such as an AME 8110 hexode plasma etchers – eighteen wafers, electo-chemical deposition of multiple parts – 200 parts, and hyperdermic rubber plungers – 500 parts in cavities.
    Cheers,
    Andy

    0
    #98889

    PB
    Participant

    Gabriel,
    I echo Andy’s sentiments. I struggle with this on a daily basis. One thing you can do is look at each cavity for the dimensional readings. (I have found that the variation from cavity to cavity will render a lower CpK but variation within cavity will render a higher CpK as you should see a tight distribution around the mean.)
    However, this mean may be either towards the Upper Spec. Limit or the Lower one. But look at each individual cavity and find the CpK. If the CpK for each cavity individually is acceptable, you can run your tool. If anyone cavity is outside, you may have to block that cavity and (may even need to re-process to see if that affects the other cavities) run the tool.
    Hope this helps.
    PB

    0
    #98896

    Gabriel
    Participant

    That’s the problem. 24 cavities x 50 parts x 4 significant characteristics = 4800 measurements. Not realistic. And this is just to run the tool. How to assess ongoing process control?

    0
    #98911

    Anonymous
    Guest

    Gabriel,
    One appraoch would be to use a stratified sampling plan based on your process engineering knowledge and use a Multi-vari chart to identify the extrema locations in the batch process. (The MV chart avoids having to re-plot X-bar and R charts with various subgroup to find the sources of variation.)
    In diffusion processes, one usually finds that the load end of the furnace runs thinner and there is more variability than the source end due to a depletion of gases down the tube and convectional heat  loss at the load end. Once you identify the sources of variation you could consider estimating a process capability based on the spread of the multivariate distribution. Obviously, the more correlation there is the better.
    Since the multi-distribution is not normal one would have to re-define Cp as ‘tolerance/ process spread. ‘ When we faced this problem in 1985 we decided to use Tukey’s psuedo sigma as an estimate of the process spread.
    As for SPC, I would consider controlling the two extrema distributions of the process, but I for the moment I can’t see how this would help calculate the process performance (SPC process capability.)
    Another approach would be to estimate the stability of the process over time is to plot the process mean and the process uniformity in the usual fashion on X-bar and R charts as described previously. In my experience, most batch processes are surprisingly stable provided that they are maintained and working correctly, although this does raise the spectre of having a short term capability the same as a long term capability as processes are usually adjusted every 15 to 30 runs or so for cleaning, or re-activation. (This is one of my objections to the shift – of course if there is no shift there is no paradigm shift – and six sigma is just a reference to +/- six sigma tolerance limits, which was well-known in MOS 3 from about 1985.)
    For what it’s worth, MOS 8, MOS 3, and MOS 2 used the Motorola Multi-vari charts in the 80’s, programmed by Mike Wolfe. This software was quite ahead of its time because along with the data it also stored all the corrective actions and calculated several process control indices, including Hotelling’s t-squared, which were shown as flags and appeared at the bottom of the chart. The chart did not use Anova for the reasons I mentioned earlier; instead it performed tests of proportions. Many engineers contributed to the development of the chart. (I gave a presentation to Motorola’s Quality Council in about 1989 – I still have the overheads.)
    Andy

    0
    #98924

    Gabriel
    Participant

    Thanks for your help Andy, but I’m miles behind that. It seems I have to study a lot before I can do (or even fully understand) what you are proposing.
    Let me give a breif introduction of the process and an idea I have, and tell me if it is too crazy.
    The mould has many cavities (let’s say 24), which are supposed to be identical but are not. Colse, but not fully identical. The material is supposed to reach each cavity in the same conditions (pressure, temperature, speed, simultaneously) but it does not. Again it is close, but not exactly the same conditions. However, the lack of equality between the cavities is pretty consistent over the time (the bigger cavity ramins the bigger one) and the way the material reaches each caviti is also pretty consistent over time (the coolest cavity ramains the coolest one).
    That makes each cavity have a particualr behaviour. I am sure that, with enough data, you could tell that the pats form that each cavity have their own average, standard deviation, and maybe distribution shape too.
    Yet, if this varaitions are small enough compared with the average part-to-part variation, they may be hidden under the noise and never be discovered. Something I do not care about too much, since I only care of variations that are of any practical significance.
    Now, I am not an expert in this process (remember I am trying to help a supplier meet what I asked him to do). But I guess that there will be a practical significant difference between the averages but not between the other characteristics of the distributions of each cavity.
    The idea:
    Find a few (let’s say 2) “typical” characteristis (Ys) that you want to monitor (because they are CTQ, good to see the “health” of the process and well measurable). These 2 characteristics should also be expected to be as independent as possible (I would not choose two concentric diameters because they will be so dependant that monitoring only one would be enough, but I could choose could be a diameter and a height)
    Run the process until it the parameters (Xs like pressure, temperature, etc…) stabilize, then make a few (let’s say 3) consecutive injections.
    Measure these characteristics in all parts (24x3x2=144, not so much).
    Plot an Xbr-R chars with 24 subgroups, where the horizontal axis is “cavity” instead of “time”. This should show significant differences between cavities. Then select the extremes cavities (will be maximum 2 characteristics x 2 extremes = 4 cavities, and minimum 2 cavities -one for each extreme-, let’s say that we have 3).
    Then make let’s say 60 injections and make an Xbar-R chart (subgroups size 3) for each selected characteristic of each selected cavity (60*3*2=360 measurements, quite a lot of work but not impossible to make it once) and use them to define the control limits. Verify that, with this ammount of info and the previous info of all the cavities, the selcted cavities still seem to be the extreme ones. Calculate Cpk for each both extreme cavity for each characteristic.
    Keep the 6 control charts for process control, taking a subgroup of size 3 for each selected cavity with a defined frequency (3*2=6 measurements each time, doable). From time to time, use the data in these control charts to calculate Cpk and Ppk and monitor the evoution.
    Sporadicaly (like after machine set-up), measure these selected characterisitcs in the other cavities (just 1 part per cavity = 24*2 = 48 measurements) and see that they are within the extreme cavities.
    For all the other characteristics (the non-selected ones), they would be measured only once in all cavities and never measured again unless there is any indication of quality problems which could be related to them.
    Of course, the Xs (presure, temperature) must be kept under control, but as they are automatic machines with PLCs and closed-loop controls to keep the process parameters on target and alarms for when thy cannot be kept on target, I don’t think there will be a problem with that. But things may happen for example with the tool or inside the small ducts that take the material to the cavities, things that can not be detected by a shift in the Xs, and hence the need to monitor the Ys too.
    This does not gives me all the confidence I would like to have (only 3 per cavity to detect differences, control limits with only 60 individual values, monitoring only 2 or 3 cavities and only 2 or 3 characteristics, subgroups of size 3 wich are not very powerful to detect shifts, ….). But it is doable and much better than what we have now: nothing.
    Am I too crazy? Is there something easier that could be done?

    0
    #98928

    Anonymous
    Guest

    Gabriel,
    At first glance your proposals seems fine …
    My only concern is if you intend to use ‘control limits’ for the exploratory run. If you do, it would seem that there are two potential problems: the first is the question the distribution across a cavity and whether you can invoke the central limit theorem with n=3, the second is that the subgroup average will probably be highly correlated from subgroup to subgroup – a kind of autocorrelation – I’m not sure what this would do to the control limits.
    With regard to the process controls, be aware that it is usual for moulders to have more than one  tool, and they often ‘adjust’ the process differently for each one. (I think it is true to say that of all fields of endeavour I’ve encountered, injection moulders are the world’s greatest tweakers.)
    Apart from the other noise variables that you’ve already identified there is the question of how quickly or slowly a particular cavity heats up and cools down, which can have a major impact on shrinkage. (If aging due to stress is a problem, do let me know.)
    Cheers,
    Andy
     
     

    0
    #98935

    bonn
    Participant

    1. You don,t have to choose 2 characteristics, only dia. is good enough for evaluation.( first you have to know the individual sizes of cavities).
    2. you can divide your mold in 4 or 6 sections depending on the distance from gate and runner(the pipe which brings the matereial to the cavity). the closer cavities and the farthest cavities ( like on the 4 corners will behave in the same manner ) In fact machine operators change setting to fill the farthest corner, they have no problem with the close ones.
    3. Ask vendor to give you 1/2 shot and 3/4 shot and full shot with all the runners and gate, if he can. This will give you the idea of mold filling and you can pre evaluate which cavities are to be focussed in the initial evaluation.
    4. once you determine the same behaviour sections, you can choose one cavity from each and do your evaluation of inputs.
    5. you should change or confirm the other cavities once in a while.
    6. in your initial evaluation include the runner length/dia/ section ( for blow holes) also which should confirm point 2.
    note : overfilling/burrs are mainly due to worn out cavities or not matching faces or too much injection pressure.
    it is a little confusing in the start but you will get it very soon.
    regards bonn.

    0
    #98936

    Gabriel
    Participant

    Andy,
    Thanks very much. I’ll take the “At first glance your proposals seems fine …” part to go to my supplier’s :-)
    Regarding your concerns, in principle there are not evident (to me) reasons why the distribution for each cavity would not be fairly normal. We are not talking about bounded characteristics. Anyway, I do not think that normality is a big issue for control charts. Any typical distribution has well above 99% of the individuals within ±3 sigmas, even if not exactly the 99.73% of the normal distribution. You know, R, S, p, u, c are never normal, and we still use ±3 sigma limts and I never heard a question against that.
    Regarding the correlation subgroup to subgroup, yes, that’s exactly the idea. I guess that the R chart will be under control, but the Xbar chart not, because I expect more variation between subgroups than within subgroups. Just like the Xbar-R charts in an r&R study. I expect to use that information to decide which ones are the extreme cavities. I don’t want to take just the one where the average is greater and the one where the average is smaller. If even these cavities were within control limits, I would have no reason to say that there is an extreme cavity at all. On the other hand, if all the ranges are in control as expected I can say (not with a lot of power) that the variabilites within cavity are comparable, and then I do not have to care about that maybe another cavity with a better (more centerd in spec) average can have a worse capability due to a larger part-to-part variation.
    I appreciate your feedback very much.
    I would also like others to jump in and give me a hand with this. You are all invited.

    0
    #98937

    Mike Carnell
    Participant

    Gabriel,
    I have a little different view of this. What you want controlled is a shot – that shot makes 24 parts. Part to part variation should not change once you optimize the mold and how the mold and the machine interface.
    The idea of multiple cavities is a homogenious part. That is a hypothesis test so that there is no variation between parts. Optimize the process that makes 24 parts and control that – not each cavity. Use some Shingo stuff to reduce process setup variation and PreControl to turn the process on.
    Just my opinion.
    Regards,
    Mike

    0
    #98938

    Gabriel
    Participant

    Wow!
    A lot of help. Thak you very very much.

    0
    #98942

    SSNewby
    Member

    Gabriel,
     
    We have several back-integrated plants that mold for our finishing plants and we have seen the best way to maintain demonstration of part-to-part uniformity is to begin with extensive mold/tool approval measuring every appropriate cut metal dimension on the mold/tool as a pre-installation approval and then at process qualification (attachment of tool to press) run a first article on each print dimensioned measurement on each cavity across multiple shots.  The process qualification itself is a DOE in which we determine optimized press temperature, screw speed (correlates to pressure), pack and in-mold residence times.   Once the tool has been inspected for metal cut and placement of gates, and the process has been qualified for the specific press and tool combination, production is turned on to run the parts.    The first production run has a full dimensional check of the part and subsequent runs have the critical dimensions, as identified on the specifications print, checked at start-up and periodically per shift as the lot progresses.
     
    From the point where we are running production on a qualified tool and press combination our critical dimensions checked at start-up and per shift are entered into an SPC program (using ASI DataMyte).   We trend and then respond to trending data as appropriate.   Issues with short shots, hoop stress, etc., resulting from tool wear, plugged cavities, materials dryness, etc., are noted and responded to.   It sounds like you are dealing with a small shop with limited resources and it is possible to separate the responsibilities, such as, your engineering can help with tool and process qualifications and the parts molder can perform basic SPC recording and trending using elementary tools and measurements from critical dimensions taken at appropriate times in the lot’s progression.  You can predetermine the issues and concerns that the parts molder needs to contact you about and/or the types and nature of the responses available for trend variance and excursions.  Frequently, as you and others have pointed out in the past, answers don’t have to be that difficult to derive because the question/problem did not have to be that complicated and convoluted.
     SSNewby

    0
    #98944

    PB
    Participant

    Gabriel,
    This thread is giving me some good info as well. I agree with Mike. What I was hinting at in my post was that once you have your process optimized, part to part variation should be consistent for each cavity. The between cavities variation will also stay consistent unless the process changes (which happens because of noise). Therefore, your initial DOE (using High/Low and center point) should focus on  3 or 4 process variables that are critical to the part processing. Once you understand the main effects and interactions (if any), you can establish a process controls on those process parameters to ensure the part to part consistency.
    Many times, the steel will not render effective results and steel (in the cavity) may need to be changed.
    PB

    0
    #98948

    Gabriel
    Participant

    SSNewby,
    Thanks for the help.
    From your description, it seems to me that you are talking of plastic injection, while this is rubber injection. Anyway, most things would be applicable for both.
    Can you rephrase “measuring every appropiate cut metal”? Is it measuring all dimesnions of all cavities on the mould (not on the part)? This is a good idea, but some measurmenets would be virtually impossible to measure without cutting the tool. Some dimension on the part also need the part to be cut, but that’s not as critical as cutting the tool :-))
    Also, what is “run a first article on each print dimensioned measurement on each cavity across multiple shots”? Do you mean to make several shoots and to measure all dimmensions of all parts? For example, I have 24 cavities, I make 10 shots, there are 240 parts, and I must measure the 20+ dimensions shown on the drawing on all 240 parts? If that is correct, the it is beyond the possibility of either our supplier or ourself.
    Also, once everything is qualified, you check and plot SPC charts for critical dimensions… of all the cavities?
    Thanks for the clarifications.

    0
    #98950

    Mike Carnell
    Participant

    PB,
    We are in complete agreement. Each cavity is not an independent event and trying to control each cavity as if it were a stand alone process will ultimately fail. Just because you have 24 cavities does not mean you have 24 processes.
    Thanks. Good luck.

    0
    #98954

    Gabriel
    Participant

    Hi Mike,
    Thanks for joining.
    This approach is fully new to me and I find it interesting. Can you give me more details?
    If I understand correctly, it is like having one “big” part (the full shot) instead of 24 independent parts. It is sound. Only that my new “big” part has 24 times as many dimensional characteristics!
    Would you give me a brief step by step implementation starting with the mould and ending with the process under control? (at least the titles of the steps?)
    You know that I always appreciate your opinions. And if you had some related material you would be willing to share with me, you have my e-mail. Of course that would be more than welcomed.

    0
    #98957

    SSNewby
    Member

    Gabriel,
     
    Typically good and on-target questions from you.  
     
    Yes.  I am referring primarily to PVC plastic injection molding.   But, as you said, I don’t believe that impacts your question.  It impacts, obviously, process methodologies, tooling and set-point parameters, but that’s not really what we are discussing.
     
    By “measuring the metal”, I was talking about tool metrology and dimensional approval done against the tooling drawing, done well in advance of the mold/tool arriving at the molding plant.   It is done, as you said, in a non-destructive manner, measuring all tooling print dimensions that can be measured.  Our tooling engineers and metrology folks are very skilled at their jobs and not much gets by them.   Their professionalism and capabilities on the front end of the process give us a very good quality assurance starting point.
     
    In talking about running a “first article on each print dimension measurements on each print dimension”, I am describing the qualification/validation process for a new tool and an existing press, existing tool and new press, or new tool and new press.  Saying multiple shots at this point usually means three shots.   We have quite a few 64-cavity tools, so this is no small task.
     
    For qualified/validated processes (tool/press/process validated) in ongoing production, we measure the critical dimensions of one part from each cavity at start of run.  The measurements are taken using gauges which have an RS232 port connection to the PC used for SPC trending.  So, the measurements are taken once and automatically entered into the SPC program.   During the production run we use go/no-go gauges, pins, etc., to verify that the samples pulled in process are still within acceptance ranges.   But, yes, we perform dimensional measurements on all critical dimensions for one part from each cavity at the beginning of each lot.   If we find that we are out of dimension on a critical dimension at the beginning of a lot, we have to go into the preceding lot and sample very heavily across the proceeding lot trying to determine when it went out of specification – if we can’t make a conservative case as to “when and why”, the preceding lot is also toast.
     
    I should probably have prefaced my comments by saying that we are in the pharmaceutical and medical device business and many of our molded parts are made to as high a precision level as possible.   But, regardless, starting with established tool dimensions, a qualified/validated molding process mating dedicated tool and press, and then using SPC at start of lots saves you many headaches in process.
     SSNewby

    0
    #98959

    Mike Carnell
    Participant

    Gabriel,
    Sorry for the slow response. You have 2 things going on. The hard tooling which will never make a part with a small part to part variation unless each cavity is the same size (statistically). You do not need to worry about all your 21 characteristics if the cavities are all the same size. The second part of the issue is how that tool gets material, delivers it to each cavity and the material characteristics themselves. I am not sure what type of molding you are doing but this is usually a machine driven process – not a lot of operator induced variation.
    If this is one of those rubber molding processes that lets operators stack rubber on top of the mold basically in a random pattern then you have a good starting point.
    When you get the tooling set up so the cavities are homogenious you run the process through optimization. Typically this is DOE’s – mostly knob variables on the machine and Variation type variables from the material. You can run something like EVOP or Steepest Ascent to get to the best location to run the process.
    You can use the stuff from Shingo (SMED) to make the tooling setup faster and more repeatable. Mistake proofing will help as well.
    When you set up the process you have the setup people use PreControl to turn the process on – that forces the setup guys to the center of the spec before they run.
    I don’t have anything written on this it is just how we do it. Sorry.

    0
    #98972

    Anonymous
    Guest

    Gabriel,
    I don’t envy you … you’ve had a lot of advice!
    Anyway, I just thought that I would point out that what we discussed is consistent with Taguchi Methods, where one  identifies a worst and best case as a ‘combined noise.’ Another approach would be to perform a stratified sampling plan based on factorial noise factors, as someone else mentioned.
    Your example is a good one and  illustrates why Taguchi advocates putting noise factors in an outer array – so that he can calculate the process mean (so-called one shot) and the process uniformity (signal-to-noise ratio) which for nominal-the-best is a measure of uniformity. It is also self-evident, that the more correlation (collinearity) there is between cavities the better, as this improves uniformity.
    Finally on the subject of the importance of finding noise factors, anyone with a statistical bent might find Roger Bohn’s paper of interest.  (Roger, who was based at Havard University at the time, compared several semicondctor facilities in the light of their ability to identify and use noise factors. He was kind enough to reference our article in Semiconductor International.)
    http://isic.ucsd.edu/papers/noiselearning.shtml
    Cheers,
    Andy

    0
    #98977

    Gabriel
    Participant

    Hi Mike,
    Yes, the mould is made in a CNC machine, and all cavities are done with the same CNC program (and in the same CNC machine and using the same turning inserts, etc.). So the variation between cavities in the mould is just due to CNC machine inherent variation. Read it is very small compared to the tolerances in the rubber part. That is, all cavities should be identical in shape (from a practical point of view), but it is not a bad advice to check (via cavities metroology) that what should be actually it is.
    However, my supplier’s experience is that even when all cavities can be considerd to be identical and the channels feeding material too, not all cavities perform the same way (they don’t get the material exactly in the same conditions of time, temperature, etc.), and hence the variation between parts of different cavities in much greater than the variation in the size of the cavities itself. I don’t know why.
    Going to the process, it is not a rubber press where the operator puts some pre-formed rubber parts inside the mould, if that was what you mean. It is a rubber injection machine, much like a plastic injection machine, only that in theis case the temperature makes the material hard instead of fluid. The temperature at which the material reaches each cavity and the temperature of each cavity itself (which are never identical) are critical to process, but they are not “knob variables”.
    Thanks for your help.

    0
    #98994

    PB
    Participant

    Gabriel,
    Not all cavities behaving identical is one of the outcomes inherent to such processes. You may have to do one of the following:
    1 – Block off certain cavities (maybe they are the peripheral ones) first and look at behaviour of the injected cavities (say you are now looking at 20 cavities). If you get similar behaviour – then you know the ones blocked off are giving you the difference.
    2 – You may have to continue to do step 1 until you see similar behaviour. If this happens, maybe you need a smaller tool with less cavities.
    You say – ‘The temperature at which the material reaches each cavity and the temperature of each cavity itself (which are never identical) are critical to process, but they are not “knob variables”‘. You may want to try ‘hot drops’ such that one drop feeds 8 cavities if that is possible on the tool. You may want to try extremes of the temperature ranges on your DOE. Also, in injection molding they add ‘flipper’ in the runners (this is a newer thing) that turns material at specific runner positions that helps maintain the material temperature. I am not sure if that can be done for rubber molding.
    These are just some suggestions.
    PB
     

    0
    #99000

    Gabriel
    Participant

    I’m not sure I understand your points 1 and 2. Would you clarify, please?
    I am not an expert in rubber inkjection (remember I’m trying to give support to our supplier). But ther rubber becomes hard with temperature, the opposite than with the plastic wich becomes fluid with the temperature and hard when it cools down. The rubber, before vulcanizing, is like a paste. It gets to the cavity just pre-heated at a not very hot temperature and it must remain in the cavity under a higher temperature and pressure during a certain time for the vulcanizing process (that’s why, I think, the cicle time is longer and hence more cavities are used than for equivalent productions in plastic). I don’t think that a hot drop would work in this context.

    0
    #99008

    PB
    Participant

    Gabriel,
    Here is a link to Dowcorning Rubber processing. You should be able to look for rubber molding, extruding, etc. and download some info. Hopefully this helps.
    http://www.dowcorning.com/content/rubber/rubberprocess/default.asp
     
    PB
     

    0
    #99011

    Mike Carnell
    Participant

    Gabriel,
    Sorry about the delay (again – in meetings in Lima).
    This is basically breaking the problem into pieces. Tooling and Process. Just because it is CNC produced does not mean they are the same. There will be some variation but as you said is it significant in relation to the tolerances.
    Process is another issue. Be careful when things are “inherent” in a process. Frequently that translates to “I couldn’t figure it out so it is just an eternal truth and we will have to control it with incense and animal sacrifices.” Heat is a knob variable. How material flows is pretty well understood and the design of sprus, vents and gates is pretty well documented. That is just a method of controlling the injection process.
    Processing the rubber is also pretty much treated as an art. I would take a pretty good look at the specifications around the material, analysis of the material (the measurement systems are a little random) and what characteristics do what.
    I am with Shree in Santiago this weekend (nobody gets up very early). We have been down this road and he understands the measurement systems well. If you want to talk please send an email to me or [email protected].
    Thanks.
    Regards,
    Mike

    0
    #99026

    Anonymous
    Guest

    The reason why it is important to identify the sources of variation is because they provide clues as to what is going on in the process. For example, the process variables that affect the process mean are generally not the same as those variables that affect uniformity. For example, raw material variation would generally affect the process mean and from run to run with time.
    Your supplier obviously knows his process well as he has correctly identified the most likely source of variation. It can be easily confirmed using a dimensional map of the part dimensions by cavity.
    If the problem is shrinkage due to a thermal variation across the tool, then the heat loss would be greatest on the periphery of the tool – Newton’s Law of Cooling. Therefore, one would expect the shrinkage to be greatest at the centre of the tool. If this is the case, the temperature variation can be confirmed by putting two thermal disclosure tabs  (edge and centre) inside each cavity and performing a ‘dummy’ run without injecting material.
    Now the question of whether or not you’ll be able to improve the process will depend on whether you have a solution for  the technological problem of heat loss, an alternative chemistry, or whether a parameter design is likely to desensitize the process to the thermal variation, which in turn depends on whether the process is non-linear or not, because you will not be able to scale the variance using the SNR.
    Once again good luck!
    Andy
     

    0
Viewing 25 posts - 1 through 25 (of 25 total)

The forum ‘General’ is closed to new topics and replies.