Today I'll talk about "The Sandy Munro Fallacy", so called because I believe Sandy is perpetuating an analysis regime that can be deeply flawed.

$tsla $tslaq
Let's start:

Sandy Munro and company tear down vehicles and do costing (cost analysis). They do a good job on that because pricing an item you can gain experience and it ultimately depends on the cost of labor and common materials.

But they fail on the second part - analysis of part design

How does something get designed? it's an open ended question but as an engineer I can tell you you will pick an approach/model that you're familiar with as a starting point, something that is known to work. Sometimes this isn't optimal and you have to go outside the box.

For example, you want to dampen the vibration of a motor. This isn't a new topic and there's many approaches, but it usually requires the use of rubber and similar materials to dampen vibrations.

Using tried and true methods lowers the risk of the design going wrong

The more outside the box you go, or if you use new methods or approaches, the more risk you run.

How do you de-risk this? You model, prototype and test.
For example, in the aircraft industry, building and testing scaled down versions in air tunnels or using modeling software

This can be done in any number of industries. For example, when I work on Antenna designs, simulation is done based on some experience, followed by prototypes and testing. It's a cycle. And for critical components it's extremely tedious and complex.

You can get a very good idea of whether your approach to something works in this way.

However, you have to test, and your testing needs to be realistic of the conditions your system will be in. It's also time consuming and can be costly

What's the issue with Sandy Munro's teardowns? Simply put, he makes quick judgement calls about a certain design item, which he may be right or very wrong on.

In Tesla's example, Sandy has no idea whether the approach Tesla took is reliable. He isn't simulating. Sample size of 1

Not even the engineer who designed it would know without data and analysis. Unless Sandy is familiar with it and understands that particular approach, his opinion may be wrong.

What's worse is that these designs need to survive extremely high production volumes.

Some issues aren't found until 3-4 issues per million, or less. So a design problem may never reveal itself even to the designer after significant analysis.

You want to tell me a guy that takes a look at the car in a few hours can say a design is reliable with no analysis?

The best example of this is Tesla's latest cooling contraption, the Octovalve.

https://www.teslarati.com/tesla-model-y-octovalve-easter-egg-teardown/

The first issue is simple - they've reduced everything, but the points of failure have also been concentrated. Now if the air conditioning cooling fails, so can the battery.

Auto OEMs are some of the most penny pinching operations in the world. When I worked at a supplier for them, it was known that a $0.50 increase could be a reason you won't be picked for their design. So why would they be "idiots" and have 2-3 different cooling systems?

Because in some cases, the cost you save upfront, you pay for 10x when the system fails.

As an example, distributing the load among two or more systems reduces load. In a split design, the AC isn't constantly running and neither is battery cooling. In a combined design, it will

So what's the risk here? That this new contraption Tesla designed could fail due to overuse. For the battery it can be catastrophic (remember NCA batteries have lower thermal runaway temperature)

In Tesla's case we know they cut corners and have quality issues

Because of that, I'm more willing to say that there's a much greater risk of Tesla's designs failing. Every Tesla engineer that I've seen talk about the culture there speaks of cutting shortcuts and doing "cool stuff"

Reliability engineering is boring, but absolutely necessary

I'm not saying all of Tesla's designs are bad. There is some innovation and changes. But these designs need to be vetted, and one way to vet is to see failure rates when a large number of products are made.

Sandy cannot predict failure so easily. No one can.

Does Tesla benefit from vertical integration and doing it in house? Yes and no. On the one hand, a custom solution is usually much better than an off the shelf. But off the shelf solutions have a benefit of greater volume (so data) and a vendor who can be much more knowledgeable.

It's easy to make something that works for a day, a week.
(You've seen the Tesla cars that stop working after 15 minutes off the lot right, right?)
But making millions of parts that work for 10 years is harder, and is why auto OEMs have are so conservative in their designs.

Tesla bulls believe that this is a unique advantage Tesla has. That can be true, as long as none of what you do screws up. But when it does, it blows up like a nuke and can completely wipe out profit for a car.

Then again, Tesla need not worry about profits. Hasn't for 17 years

I'll add this - Complex systems, for obvious reasons tend to fail more easily.

Remember that Tesla took a $1 rain sensor and replaced it with a failing computer vision approach that sucks.

Being complex may appear beautiful, but it may not work well.