Spot Weld Optimization using Data Science

Vinaya:
We have Mohan with us to talk about this. Hello, Mohan. What initial challenge that the customer encountered that could be helped with our solution?

Mohan:
We had already done a pilot with this particular client.
So the challenges in this case was, ‘How do we formulate the problem?’
For that we wanted to understand the existing process.
Once we understood what they are doing, what kind of data is available, what kind of digital data we can use for our use case, we started experimenting with that.
They were a little uncomfortable that the geometrical data was missing. But then it was not needed for our modeling. We had to convince them a bit. Apart from that it was a good start in this project having already proven that the concept works in the earlier project.

Vinaya:
What process did you follow during implementation?

Mohan:
We understood the existing process. They were doing several iterations. In each iteration, they were trying to reduce the number of spot welds one by one though it was not as simple as that. Every iteration they will try to do it and then decide the number of spot welds required is more to provide the same strength to the joints. So, we looked at how it is changing the number of spot welds in different areas whether the required strengths, targets are reducing or increasing. We collected some data, sample data, from their existing processes, existing stimulations what they had done. Then we started applying it, using our own tools. We tried to do it in step by step showing some samples which fills the design space and measuring it for various things. In fact, our target here was noise vibration harshness, durability, and safety. We covered all these in sequence and then solved it.

Vinaya:
What road blocks for implementation did you help the client overcome?

Mohan:
The number of stimulations they had to do, they were thinking it would be too many. But then we assured them saying that we will do in an incremental way of giving them the number of samples to stimulate in their existing systems. So, we took a very small percentage of what they were actually doing and then built the model on that. Looked at what parameters are really impacting the targets. Then we focused only on those. We then gave them the next set of samples. That way we could reduce their samples by almost 60% initially. We probably added 10% to give a better solution. The challenge they were thinking that the amount of time they spend time providing the data was reduced to half.

Vinaya:
What kind of benefits did the client enjoy with the solution?

Mohan:
First and foremost, they could save about 9% spot welds for each car. That resulted in approximately 13 million $ for the number of cars they produced for that model. In addition, it saved about 40% in terms of manpower effort.

Vinaya:
Was there anything about the implementation or the results that positively surprised you?

Mohan:
Definitely. We were thinking it was going to be complex, non-linear system. But, it resulted to be more linear. That is one part. The second part, we ourselves couldn’t believe that we were able to save, when they actually looked at the total amount of money they could save was crossing millions and not only a million but 13 million $. So, that was great.

Vinaya:
Thank you, Mohan. Thank you for watching.