Diagnosing Unexplained Product Failures

Sometimes we’re called on to act as an independent third party for clients who’ve completed their internal design process and need help analyzing an unexplained failure. Diagnosing problems in failing products that have already been released for manufacturing can be a tricky business. First, no one wants to hear that the tool needs to be scrapped. Second, if the failure is not well understood, no one wants to pay to thicken a wall or add a rib without knowing if it will definitely solve the problem.

An example from recent work: A client came to us with a consumer product that was breaking in the field after it was vetted and released to manufacturing. They assumed they were going to need to make a design change and probably a tooling change as well.

In this particular case we had three sets of samples: a late generation prototype that had been extensively tested and never failed; a first article part that had been approved; and a production-run set of parts that was failing badly.

Getting to the Root Cause

Throwing Band-Aids at a problem can be the knee-jerk reaction when you have failing inventory piling up in a warehouse, and in fact our client had already asked the molder to beef up some bosses and external walls. Band-Aids, of course, don’t solve problems — in fact they tend to mask the root cause. Getting to the true root cause is critical to long-term quality and reliability. But, of course, time is of the essence.

With samples in hand it quickly became clear to us that a thicker boss wasn’t going to solve the problem. The first clue was the nature of the failure. Just by manipulating the parts in our hands the plastic parts developed brittle cracks — not just in the thin sections, but also through the thick. Brittle failure is not expected behavior for a part specified as polycarbonate.

To try to quantify that hunch we mounted the parts in a fixture on a tensile test stand (as reenacted below) to compare the material properties of the three different generations of parts. Obviously we didn’t have a standard tensile test specimen of each material, but by plotting the load versus displacement of each generation, it immediately became clear they were not all created equal. There was a 2:1 ratio between the load at failure of generation 1, compared to generation 3. This led us to believe that, at minimum, the material was not what had been specified. When coupled with the nature of the brittle cracks, we began to suspect that something had happened during processing.

The force gauge in action.

The force gauge in action.

The force reaction being simultaneously graphed.

The force reaction being simultaneously graphed.

Developing a Testing Protocol

Despite this hunch, time was still our enemy. In parallel with sending the material out for analysis we also developed a testing protocol to allow the manufacturer to test product coming off of the line. We knew they could produce good parts — they already had — but in the short term the parts needed to be qualified prior to shipping.

By rapidly developing a testing protocol that good parts could pass and bad parts would fail catastrophically (a go/no go test), we intended to enable production while looking for the real issue. We verified this testing protocol with a sample size of 50 to make sure we would get a good distribution.

Testing the Parts

While the testing protocol was being developed, the material samples from the three generations of parts were out being tested at the lab. The first tests confirmed the material types.

We had three generations of parts:

1. Gen 1 was Nylon
2. Gen 2 was PC (and passed testing)
3. Gen 3 was PC (and failed testing miserably)

Next, a melt flow test was conducted to determine if the mechanical properties were compromised. The faster it flows, the more degradation has occurred (the chains are shorter). This test revealed a problem. Both nylon and PC (the two materials in question) are highly sensitive to moisture content in the base resin. The resin needs to be dried to approximately .02% water content. If it’s not, the water breaks down the molecular chains, making them shorter and lowering the mechanical properties.

The way to determine this after the fact is to do a melt flow test where the polymer is heated and the flow of the molten material is observed. The faster it flows, the more degradation has occurred (the chains are shorter). What was interesting about the testing was that the Gen 2 PC (which passed testing) was actually quite compromised relative to what the material properties should have been.

Mystery Solved

PC cannot be consistently molded without desiccant driers — something that few Chinese molders have or use. Most Chinese vendors use hot air dryers and in some cases (on hot humid days) this actually adds water to the resin. The result is that parts molded on hot humid days have the potential to be really bad, and the next batch molded on a cold dry day is good … a supply chain nightmare!

A couple of additional comments:

1. It turns out that ABS is much less susceptible to water/drying needs. This would seem like a natural substitute for many applications, but keep in mind that ABS is much more susceptible to UV degradation. A UV stabilizer only delays the inevitable.
2. ASA is a good choice — it is similar to ABS but with better UV stability.
3. A PC/ABS blend is not much help, because it tends to be 70% PC and therefore need to be well dried.
4. Heat degradation (time and temperature) can also occur during injection molding with similar results.

This was an interesting and actionable result. Tighter controls were placed around the drying requirements and, to make a long story shorter, subsequent parts met the necessary mechanical requirements. This was a satisfying ending to a serious problem. Coincidentally a week later a completely different client and molder came to us with parts that were mysteriously underperforming their mechanical design points….