Customer success stories with MiViA

I’m very pleased that you’ve taken the time to join us today. Before we begin the interview, it would be lovely if you could briefly introduce yourself — your role at the company and what your responsibilities in that role actually are. Yes, I’m Philipp Wehnert, 33 years old, a qualified materials and testing technician at Kugel- und Rollenlagerwerk Leipzig GmbH. I head up the hardening shop, the materials laboratory, and the non-destructive testing department.

I oversee around 15 employees and two colleagues in the office, who work with materials on a daily basis — predominantly steel, both in the hardened and soft condition. If we turn now to microstructure analysis and metallography — what are the key challenges you face in that area? Well, subjectivity is always a major challenge. Many people are familiar with the basic microstructures — ferrite, pearlite — but at the very latest when it comes to hardened microstructures, distinguishing between martensite and bainite, or identifying artefacts — the question always arises: is this really the case?

And what can we do as a company to obtain consistent results from every colleague? " And that’s how the contact came about, with the idea of removing — or at least reducing — subjectivity from the laboratory, to explore new approaches that might also be established across the industry as a whole. If we stay with the topic of subjectivity — how does subjectivity actually affect your results in microstructure analysis? Well, there are various auxiliary tools within the software — for instance, grey-value analysis.

And if the microstructure image has stronger contrast, the grey tones will naturally differ. Particularly with very fine-acicular martensitic microstructures, it becomes rather difficult to determine the percentage with precision. In those cases we can have a percentage difference of 3 to 5 per cent, depending on the individual and how they’re performing on the day. There are certain percentage thresholds — 5 per cent being a limit, for example — where, with certain parameters, one person might assess it as above 5%, whilst another would say it isn’t the case.

Especially apprentices always benefit from additional guidance — and that’s precisely where MiViA can provide that support. Was there anything in particular that prompted you to try MiViA? You mentioned apprentice training. The primary focus was the fact that it’s new and innovative — it simply doesn’t exist yet.

That means there’s an opportunity to help shape and develop it, which is very interesting. Rather than having a ready-made tool that’s already been tailored for someone else, we’ve been involved in the development from the outset. And there’s a vision — the challenges exist and are shared by us as a manufacturer as well as by our customers. There’s a grey area that one simply accepts — whether the microstructure is fine-acicular or acicular is debatable.

" — that’s what I’d take away from this. Were there any other alternatives that you considered? No. None that I’m aware of.

Neither from microscope manufacturers nor from software manufacturers was there anything similar that could have been considered. What specific advantages have you identified from using MiViA? It’s faster. We did a trial run, and it’s clear that theoretically, if everything ran optimally, we could realistically manage three to five minutes for three samples.

Three to five minutes for three samples. And that’s quite significant. In busy periods with a high sample throughput in a short time — particularly for routine tests where only a single value needs to be determined, and you don’t need to examine the entire microstructure as you would with a failure analysis — it’s really just about confirming that a value is present. For grain size, for example — what grain size number applies — if I run a scan and have, let’s say, 20 images, I can upload them all at once and have them analysed in a fraction of the time it would take if a member of staff were to do it.

What we’re seeing now is that customers‘ demands and technical delivery conditions are continuously tightening. Quality requirements keep growing, and this naturally means that analyses must become ever more precise — and features that previously no one had any interest in suddenly become necessary. And with MiViA, we can simply upload images to identify certain characteristics and have them evaluated — we’re faster than if you were to arrange it with customers separately. Both we and the customer could run the same analyses via your company.

We also have situations where we carry out microstructure evaluations and the customer does it differently at their microscope. You then need to agree on the magnification, the area — these are naturally things the AI would need to take into account equally. But you could share an image, send it, and both the customer and we would get the same result. Would you recommend MiViA to other companies, and if so, why?

Yes, fundamentally — because it can essentially serve as a standard. You send the same or similar images to an external party, you get a value back, and there’s no bias. That means every company that evaluates microstructure images in any way — with or without a specific problem behind it — will be able to benefit. Because it’s smarter and also more effective in many cases.

Every hardening shop — and I’m somewhat connected in that network — would find it worthwhile, because there are occasional disagreements about microstructures. And I could also imagine technical colleges, for instance — they could potentially save on teaching staff when analysing microstructures, because trainees could carry out their evaluations using the microscope and the software accordingly. So which industry would probably benefit most from MiViA? The metals industry, definitely — the metal-processing industry.

I’d say it spans from the steelworks right through to the end consumer who only carries out incoming goods inspection on the finished product — but it actually starts right back in primary metallurgy. Semiconductors would certainly apply too — anywhere you capture an image and wish to determine characteristics, there’s potential. Perhaps also in medicine — though that’s not my area of expertise. Are there perhaps any features or extensions you’d like to see from us in the future?

Yes, of course. The vision is that all microstructure images one uploads can be detected in some way. Ideally, mixed microstructures — having certain proportions of pearlite, martensite, bainite perhaps, retained austenite content — all comprehensively analysed and determined through a single tool. Thank you very much for your time and for the valuable insights.

Is there anything else you’d like to add? I look forward to the future. There’s still a journey ahead of us that we intend to pursue. The next project is already in the starting blocks, and it remains exciting.

Well, thank you very much and have a lovely day. Likewise, thank you.

Right, then, a very warm welcome to our today’s — yes, today’s interview. Thank you very much for taking the time to join us. Of course. To speak with us today, and first of all I’d suggest you introduce yourself briefly and explain your role within the company.

Yes, with pleasure. My name is Peter Maas, and I work for the company VDM Metals, where I serve as Central Laboratory Manager. VDM Metals is a large, internationally active manufacturer of semi-finished products made from high-performance metallic alloys, predominantly in the nickel-base sector. We also work with zirconium and some speciality steels.

We have a very extensive laboratory infrastructure across several sites in Germany. The full range of classical destructive materials testing — mechanical, physical, chemical, spectrometric, corrosion, and of course metallography as well. Very good. We’d now like to delve a little deeper into microstructure analysis and would like to understand briefly what the workflow looked like before you integrated MiViA — shall we say — and what a typical microstructure analysis looked like in your organisation.

Perhaps you encountered certain challenges — perhaps you could outline those briefly. Well, the process runs classically — or shall I say typically — in such a way that we don’t actually have a real standard. Because everything we produce is ultimately bespoke with customers. We have relatively few products that we manufacture for stock and then sell through our own service centre.

As a rule, we don’t begin melting — that is to say, we melt our materials entirely in-house. However, we only do so when there is a dedicated customer order behind it. This means that at the moment we begin melting, we already know with reasonable precision what requirements the material will need to meet. A so-called inspection plan is then produced, and our colleagues from technical clarification and the quality department specify to the laboratories the tests that are to be carried out within those laboratories.

The applicable test standards are referenced, and then of course the target values — such as the permissible grain size, for instance, and the cleanliness grade classifications — are defined. The production facilities manufacture the material and cut a sample from a defined point in the manufacturing process. This passes first through the sample preparation department and then into the metallography laboratories. Of which we actually have several.

And then the colleagues there review the inspection plan accordingly. What exactly needs to be examined here? The examination is then carried out. They also prepare the samples themselves.

That is also done within the laboratories. The results are analysed and reported back into our central quality management system, and at that point the order is essentially complete for the laboratories. Then it can of course occur — depending on the nature of the order and the customer — that follow-up queries arise from the quality department, from research, or from process development, who then request that something be analysed in greater detail. And I would say things become particularly interesting in the context of, for example, complaint handling.

The customer may have carried out their own inspection, perhaps as part of incoming goods checks, or possibly in the context of a damage incident, and then contacts us to say there is presumably something not in order with the material, and they have queries about it. Mm-hm. Yes, quite a broad range of… But I would have estimated that within the metallography laboratories it is probably around 90 to 95 per cent release-relevant testing, with some work for research and complaint handling — which is, thankfully, one must say, a very small proportion of the overall activity.

Quite so. Were there particular bottlenecks or issues that perhaps prompted you to look for a new solution? Well, our bottleneck shifts within the laboratories. It depends very much on the current order situation and the nature of the orders.

It can happen that — for whatever reason — there are orders for which no metallography is required at all. When we have a large volume of that type of order, the laboratories are generally well occupied, but they are not a bottleneck. And then of course there are orders — particularly from the aerospace sector, for example — involving highly demanding and extensive testing requirements. When a significant volume of that work comes into the laboratories, it very quickly becomes a bottleneck.

That is of course quite frustrating, as throughput times are always a concern. With aerospace orders it may not be quite as critical as with other orders, but even there we have a clear interest in ensuring that things move through quickly. As for quality problems — I’m pleased to say that we do not have qualitative issues, thankfully. Our staff are appropriately trained — in most cases trained in-house — and highly qualified.

That is to say, they are absolutely capable and competent to „` to carry out the inspection. We are also already seeing a rather interesting trend, in that particularly within the aerospace sector, quite significant volume increases are heading our way. A factor of approximately five is what we are anticipating. And the analyses are extremely involved, and it is becoming apparent that we shall simply not be able to manage that.

I cannot simply take on more staff — that is, firstly, extremely costly, and secondly, I would first need to find suitably qualified personnel at all. We do run our own training programme, but not at that scale — and let us say that even if money were no object whatsoever and I could expand the workforce substantially, I quite simply do not currently have the physical space to seat the people I would need to take on. We are in office accommodation, at the end of the day. Which means that headcount growth is not a viable path for us.

Okay, but have there already been noticeable — shall we say — consequences from the challenges that you are now beginning to feel, any impact on results, so to speak? No, not actually. But it is also fairly clear within our organisation — shall I say — it is demanded of us that we say, the… The results must be reliable, they must be correct, everything must be valid — and we do not sacrifice accuracy in favour of speed.

That is very well understood across the organisation and is accepted, because quality is of course our unique selling point, and one we trade on quite effectively. I mean, if we were now to entertain the idea of saying, well, let us be a little less precise in exchange for twice the speed — thankfully, that idea would occur to no one. And that is also accepted within the quality department, the laboratories, and within the sales team — and by anyone who might occasionally try to rattle the cage — it is absolutely accepted that they say: you take the time you need for this, so that particularly in such sensitive areas we do not end up — perhaps with some delay — having very awkward conversations. It would be an absolute worst case if someone were to come along and demonstrate that the inspections we had carried out were not correct.

Yes, that is entirely understandable at this point. Was there ultimately anything that prompted you to give MiViA a try? Were there specific features or benefits that you identified? Well, in fact it is what we have already mentioned — we expect that tools such as this will enable us to work faster.

That in the same amount of time, with the same personnel, we can simply achieve more — and perhaps not necessarily in the sense that one would say these highly critical aerospace analyses are something we would now have carried out by AI — supported by AI, yes; assisted by AI, yes. The thinking tends more towards saying, well, there are also many routine analyses — in inverted commas — where one needs a grain size measurement and not much more than that. The idea being that in this area, AI could perhaps genuinely take a great deal of the workload off us, so that with those valuable personnel resources we can simply free up time for the analyses that one would perhaps then also prefer to have carried out by a human. In regulated sectors, you cannot simply tell an aerospace client that AI will handle everything from tomorrow.

They will require comprehensive validation. One does need to tread carefully there. But if one says, right, standard oil and gas analysis or similar — perhaps AI can handle 80 per cent of that. The materials tester takes another look, puts a tick next to it, and on we go.

We do expect there to be quite a considerable gain for us there — simply in terms of the volume that we are able to process. Okay, good. Were there any alternatives that you also considered? No, not actually.

MiViA came to my attention via LinkedIn. You are quite active there. And I had a playbook sent through to me. And then, not entirely unexpectedly, contact was made.

No, that was actually the first — the first AI application that caught my attention there, and we said, let us simply give it a go. And had it not worked out this time, we would have had a look at what else might be available — but for now, everything here is running along rather nicely. And so we are happy to continue testing it. We did look at it a few years ago, though at that point I was not yet in a position of responsibility.

My colleagues had already taken a look at it — other applications, that is. I am not entirely sure what they looked at. But it seems there were quite considerable problems at the time with identifying twins, identifying grain boundaries, and filtering out artefacts such as scratches and dust — that sort of thing. That did not work well, and with that, the topic here was essentially settled.

That was, however, at least five years ago, I believe, and I thought to myself, well, perhaps there have also been some developments in the technology — so let us simply try it again. How has your workflow changed so far since you began using MiViA? Are there already any concrete benefits that you have perhaps identified? Well, what we are seeing in practice is, of course, that — what is particularly interesting is that the reports generated provide quite considerable added value in terms of information.

So, I would say, the histograms showing the distribution of grain size classes and so forth — these are things that in our previous way of working we either did not capture at all, or only did so in special cases when explicitly required, or when there was some kind of — I don’t know — formally agreed special arrangement with a client or with research. But that was always something we tried to avoid, because it is simply enormously time-consuming and resource-intensive. And here you essentially receive all of that at no additional cost. And that was something we had not anticipated at all, I must say.

It was not the reason we began looking at MiViA — the idea of extracting even more value from the data we already generate. Over the past few days, it has actually begun to shift, and we find ourselves saying that this is genuinely fascinating. We are currently looking at how we might actually put this to use, but it is already clear that a great many departments within VDM Metals have an interest in this kind of capability — at the very least Research, Process Development, and the Quality teams, to whom you show this and they immediately say, good heavens, that is remarkable. And then, first of all, working out what exactly we do with it now.

Any number of possibilities come to mind straight away. These are, of course, opportunities to think in directions that we have not previously explored at all. And that is genuinely exciting. Once we manage to bring this into practical application, it will represent a very real step forward.

Conventionally, one determines a grain size — a single value in micrometres or a class — and that, ultimately, is all there is to it. We are now learning, through our aerospace qualification processes, that this is insufficient in many cases. That is to say, the client then wants to know: what is the grain size at the core, in the transition zone, and at the surface? Is it homogeneous?

And so on — all of which can of course also be done entirely manually. But the effort involved escalates very rapidly and very significantly. And now that we are engaging more intensively with aerospace requirements, we are realising that with the conventional approach we will simply never be able to scale beyond a certain volume threshold. without tools like this, it becomes very, very difficult to grow in this area at all.

It is almost by coincidence that this realisation has arrived simultaneously with the MiViA trial here — making it quite a compelling scenario, certainly. Are there, beyond what you have just described, perhaps some concrete examples in terms of time savings or objectivity? I appreciate you are currently in the trial phase, but perhaps there are already indicators where you can see that things are moving in the right direction. The quality is already impressive.

The system has not yet been thoroughly trained on our specific materials. I fed in a test image fairly early on — taken from a job that I know particularly well from my own history here. And the result was, shall we say, 98 per cent accurate. As a first-shot result, that is quite extraordinary.

And that is proving consistent. We are testing it across a very wide range of materials, across different semi-finished product forms and dimensional ranges. Including cases where we know that the microstructure characteristics vary considerably. And the AI handles a remarkably broad range of these out of the box, very capably.

This is not something we are actively using yet, but it is clear that there is substantial potential here. This capability — uploading an image and having it analysed — is superb. We would actually like to develop this further, so that we might upload five, six, or seven images relating to a single job and receive a consolidated overall evaluation. These are, however, the sorts of things we discuss with you in our regular meetings.

If we were able to get that working, then — I cannot yet put a number on it, we are simply not far enough along in the trial to do so — but it is evident that the impact will be very considerable. Yes, quite. And that brings us to the topic of functionality. So you would genuinely be looking at a much broader-scale evaluation of specific areas as well.

Yes, that makes very little sense — the idea of restricting it to a single area in some way. There are areas where we have not yet tested it at all. These include areas where we also produce metallic powders for additive manufacturing — so that one might discuss morphology analysis of powders and particles, or topics such as purity levels — which, I now see, I do not believe exists yet as a module. Crack analysis — that is, surface damage arising from the production process — these are all things we would very much like to have.

But of course — I mean, ultimately the AI does not care either way, whether I now insert metallography or a micrograph of a sheet metal sample from strip or from wire. If we use this, it only makes sense if it is used throughout. Those were the topics that you might wish to see addressed in the future. But perhaps, or to put it another way, which companies and industries would benefit from MiViA if they were to use the software?

Well, I can only view this from the VDM Metals perspective. For us, ultimately, every customer who receives metallography as part of their sampling process — customers don’t generally choose that themselves; they order in accordance with various international standard specifications, such as ASTM. Or a TÜV specification, for example. And as soon as microstructure analysis is a relevant factor — and it is in a great many cases — it becomes immediately applicable.

The oil and gas sector is picking up somewhat at the moment — we can see that in their requirements. They are now beginning to demand considerably more, and that is where it becomes particularly interesting when discussing the higher information density of the images. I see the benefit here for virtually all of our customers, across the board. Although, as I mentioned, one must be mindful in highly regulated industries — they may be somewhat put off if you say that AI handles everything without human involvement, which won’t go down particularly well — but that is fine.

The final ten per cent is still done by hand. That part remains with the human, and the AI supports the remainder and frees up the time we need to focus on such topics. Yes, I can understand that. Would you also recommend it, as we do, to other companies?

And if so, why? Well, I would say it is working extremely well. It is genuinely impressive how well it functions. And I see a very significant benefit and a clear competitive advantage.

Which is why, to be perfectly honest, out of self-interest I would not want to recommend it to our competitors. But of course, I mean, anyone who works with micrograph analysis will benefit from it. I should say, VDM Metals is in a relatively comfortable position in that we can afford to invest considerably in this area. If I consider, for instance, the Sauerland region — there are quite a few small and medium-sized enterprises here.

I can imagine that they will simply no longer be able to avoid such tools in the future. And these companies typically do not have fully qualified metallographers. Nor are they accredited to ISO 17025, or anything of that nature. When they are confronted with such market requirements, they simply will not be able to meet them.

In that case, I believe the appeal is less about time savings and more about having the capability at all. I see, understood. And that brings us to the end of our interview today. Is there anything further you would like to add or contribute here?

Well, I have naturally positioned this topic — it is placed within VDM Metals, and a great many people there are interested in it, and our colleagues from metallography have a great many ideas, which we incorporate into our regular meetings when we speak with your colleagues. It is very encouraging that they genuinely take that on board. You can see firsthand what agile software development looks like — where we say, this is a feature we would like to have. Right, understood, we’ll do it — and then two weeks later it has been implemented, incorporated, particularly with regard to upload functionality and the like, which then significantly enhances usability for us once again.

For now, we are very positive about that. And one reads a great deal about it. Now one gets to see it firsthand. That is rather impressive.

We are not accustomed to things moving this quickly in all contexts. We are a relatively small team, but one can see that the people involved are genuinely highly motivated, and it is a real pleasure. And now we shall see. Our trial period runs for another three or four weeks, I believe.

And then after the Easter holidays we will need to discuss how to proceed. But it is already clear to me that we wish to continue with this, in whatever form that may take. The challenge now is that we will of course also need to determine how we integrate this into our workflow here. So we build out our systems and integrate such a solution.

At present, of course, it is all still somewhat manual and a little rough around the edges, and I believe you also have ideas about how that could be approached. But these are the conversations we will need to have next. Indeed. Very good.

Many thanks for the conversation.