Tip of the Day 133: Alarm Stability Dialog and “N” Shots

Part 1: Recommended Number of Shots for Stability

Release 8.8 of the eDART™’s Alarm Settings tool turns on the feature that checks for stability before selecting alarm levels. However the number of shots over which it checks is set to 4 as the default (since Release 8.1). This is the same number as the number of shots in the “Set Suggested Levels” dialog.

Checking stability or setting suggested levels over only 4 shots does not make much sense. We recommend that you use 20 shots. This provides the Alarm Settings tool with enough data to make a better judgment.

Part 2: What the Stability Check Is, Why and How to Use It

Tip # 77 explains how the initial alarm levels that the eDART puts in are based on standard deviation over a number of shots. This means that it assumes that this is a “normal” process and it makes a good guess based on the normal variation it sees.

We found that customers often set alarms after a very few shots after startup. The process would not be stable and the variation would look huge to the eDART. So it would pick wild alarm levels that made no sense at all. The stability check prevents that.

When you add an alarm (or a group of them) the stability checker watches all of them to see if the variation is random or not. Non-random variation includes trends, cycles, spikes, drop-outs or steps. If any of these things have been happening during the number of shots selected (the 20 recommended above) then the stability will be low. As the process becomes stable the number climbs. When it reaches 97% (for the least stable alarm selected) the eDART computes the alarm levels and fills them in for the new alarms.

You can see that there is a button that lets you set the alarms even if the process is not stable. If you do this at a very low stability number the alarm levels will be very wide relative to a normal process. Some processes may not quite reach 97% stability. In those cases, if you see the stability number level off at some slightly lower level you could click “Set Alarms Immediately” to carry on.

If the stability number refuses to rise then something is happening in the process. Put the values in question on the Summary Graph and look for the causes. For example a cyclic hot runner control (un-tuned PID loop) can cause an oscillation in the values. If you consider this normal you can set the alarms anyway. Better, get the problem fixed first.

If you wish to watch the whole process of setting an alarm with stabilization there is a 5 minute video here (the start controls are at the bottom of the image).

[youtube=3Z3zy0HLqC0]

Not much fast action drama but you can slide forward to speed it up. This shows the addition of a “Minimum / Temp @ Mid Cavity #In” value to the alarm settings and how the stability number rises until the alarm gets set with reasonable high and low levels.

Tip of the Day 127: Failsafe (Good Sorting Output) Required for Missing Values

Recall from tips of the day #’s 1 and 2 that you should use Sorting Output / Good Control for all applications in which you are not doing 3-way sorting. This technique ensures that if anything is wrong or missing the parts will not be shipped. Here is an additional point to reinforce the importance of this method.

When the eDART™ cannot compute a summary value that is used in the alarm settings then it places a question mark on the value and calls it a warning (the word “Warn”). This turns the Good Output OFF but does NOT turn the Reject output on.

Why would the eDART not compute a value? One of the most common reasons is that the sensor cable was not plugged in making the inputs invalid due to a “No Reply”. It may also not compute a value if the necessary data to compute it are not found. For example, if you set an alarm on Process Time / Cavity Fill and the pressure in the cavity never reaches the fill threshold then the eDART cannot compute a fill time (see tips # 19 and 99). Rather than filling in an arbitrary number the Alarm Settings tool records it as a warning and the Diverter Controls counts it as “Suspect”.

All of the above cases above produce a “Suspect” part that turns off both Good and Reject outputs.

Therefore, if you want no suspect parts shipped you must only use the Good output for sorting.

Tip of the Day 126: Was the part accepted or not?

As I mentioned in tip #119, keeping all the data can be very valuable. If you are sorting parts for quality a part of the summary data is a number that indicates whether the Diverter Controls output was “Good” or not. Specifically, this value is called “Status / Part Alarm Value”. You can add it to the Analyzer’s or eDART‘s summary graph. It will look like simple square up-and down values. The number “1.0” is when the part went out with the diverter in the “Good” position. The number “3.0” is recorded when the diverter is in the “Reject” position and “2.0” for neither.

Here is an example showing the part alarm value in black.

It is important to understand that this value indicates the position of the Sorting Output contact closures for any reason. If you press the Diverter Control’s “Don’t Sort” and the parts are discarded, even without alarms, then the Status / Part Alarm Value is recorded as 3. If you have “Reject After Down” set up to throw away the first 7 parts after startup then those parts are marked as 3, even if the alarms show them good.

Thus if your customer wonders whether or not those parts got downstream into his shipments you can show him the data that proves certain parts were not kept.

Recall, though, an important point from tip #60, “What Happens After the Sorting Output?”: The data will tell you what the relay contact closure did. The down-stream mechanical and human processes must also be robust to prove that a part actually did get separated that was not good. For example an open conveyor or box for parts going to the customer is an opportunity for someone to pick up a part off the floor and set it there by accident.

Tip of the Day 117: Track QC Parts with the Part Sampling Tool

On a recent consulting trip we were able to track a problem with a part by looking at summary data over nearly a 2 year period. We found an area where the pressure cycle integrals had run high. They had sampled parts all through the period and still kept them (being a medical part). But we had a problem trying to determine which parts were sampled when. The customer had batch numbers and the molder had theirs (though all named by date only and not time).

Here is a suggestion:

Every shot removed from the machine and sent out for QC or specification testing should be taken with the “Part Sampling” tool on the eDART.

No exceptions. If a part is removed from the production line for any reason (other than scrap) it should be marked.

This way, whenever a measuring question arises you can find out exactly when those parts were made, what was happening in the cavity or how the machine was set up at the time. In this graph notes were only taken on a sporadic basis: experiments, tool maintenance etc.

Here Part Sampling recorded notes for each “group” (sample) on a regular basis. The parts that were made during the suspicious period are documented by sample or batch number and the data can be matched to them.

This greatly reduces the time to find a root causes. You do not need to go through the whole process of entering part measurements into the analyzer and doing correlations. This suggestion is simply for tracking for future problem solving.

Starting and Using Part Sampling

For those who have not used it, the Part Sampling tool can be started from the Summary Graph on the top menu bar shown here…

You can also use the Architect to drag it on to the Main Menu (on Architect’s System tab) if you want it running only when sampling.

Or you could drag it onto “Molds” if you want it to be always available on every job (and can’t be stopped).

This graphic describes how to use Part Sampling in this way:

Tip of the Day 103: The Lost Sensor and the Good Output

A Fable

The Good Output was known by all in the eDART kingdom to be a fully trustworthy and reliable fellow. He spoke to the robot and would only tell him when a part was good. So the robot never needed to guess. Besides, he only listened to what the Alarm Settings told him and would only say “Good” to the robot when told. Alarm Settings was a very smart guy. He knew every important sensor, listened to each one and made careful judgments about the parts.

But the king decided not to hire the Good Output, being that Good was a little more expensive than Reject. Also the king did not like his philosophy. Instead he hired the straight talking Reject Output because that is what his robot liked to hear anyway. It was easy for the kings men to get the robot and Reject to talk to each other.

One day a sensor got lost. No one knows how. He just stopped communicating (I think his cable was unplugged; happens to us all). Alarm Settings noticed it and marked him “Suspect”. With the sensor silent he didn’t know what was going on, whether the parts were good or bad. Alarm Settings told the Reject Output that he did not know the answer any more. But Reject said “That’s ok. At least you are not telling me they are rejects.” So he never turned his output on. He’s not a bad chap. Just a little limited.

Sadly, a bad part was made that was not full. But no one knew about it. Alarm Settings was still unsure and so could not tell Reject that there was a reject. So Reject did not tell the robot to reject that part and an evil part escaped into the world. As far as we know that part is still at large and the king sits with trepidation waiting for news of the mischief the bad part might do.

So the king fired Reject and hired Good instead. After that Good would only let the robot keep the parts if all the sensors were alive and well. But it was a sad and costly mistake.

Moral:

Always use the Sorting Output / Good Control signal if you want to be sure no un-measured parts escape (tip #1)

If your equipment will not accept a “Good” signal then use the Sorting Output / Good Control anyway and wire the equipment to the “Normally Closed” contact (tip #2). You only need Sorting Output / Reject Control for 3-way sorting.

Tip of the Day 69: Multiple Sorting Outputs on the eDART – Part 2

Tip of the Day 68: Multiple Sorting Outputs on the eDART – Part 1

Tip of the Day 65: The “Race” to sort on Cycle Time

The previous tips develop the idea of the “race condition” between the time the eDART puts out the sorting signal and the time the automation picks it up. To prevent a race there are ways to force the alarms to be computed early by setting a sorter integration limit or using end of screw run as the integration limit.

However there remain three values of type Sequence Time that the eDART cannot compute until the cycle ends: Cycle Time, Mold Clamped and Plastic Cooling. This is because the eDART does not know the length of the cycle until it ends. With Mold Clamped in use (defined in Sensor Locations and working) the end of the cycle is when Mold Clamped goes off. This is also assumed to be when plastic cooling ends.

The eDART software is designed to sort of “change its mind” at the end of the cycle if you set a sorting alarm on one of these values. It is possible that the alarms on other values (e.g. cavity pressure or temperature) will indicate a good part but the cycle time will be too long. Some customers have set an alarm on Sequence Time / Cycle Time in order to prevent a long cycle from delivering a part with degraded material (due to residence time). In this case the “Good” output may go on at the integration limit and then off again at the end of Mold Clamped as shown below.

Furthermore the computation of alarm and change of output at the end happens very fast. This is because the eDART only needs to compute and compare three times. “This can take only fractions of a thousandth of a second. So the change in the “Good” output can happen simultaneously with the mold clamped signal going off. If the robot is looking for the signal at or before mold clamped then it will miss the change from “Good” to not good.”

Another Method for Rejecting Long Residence Time Parts:

Suppose your automation cannot react properly to these changes in alarms (or the race between Mold Clamped off and the change in Good output). The eDART now contains a feature in the Diverter Controls (as of Release 8.0) that lets you tell the eDART to reject a certain number of parts after down.

The eDART assumes the job is down if 30 seconds elapse after the time the next cycle should have started. You can use this method instead of setting a sorting alarm on Cycle Time. This will reject as many parts as necessary to “clean” the barrel of degraded material.

Tip of the Day 64: Time to Compute Alarms

Tip of the Day 63: When does your Robot Listen?