Closing the Loop Between Product Design and Manufacturing

By Harri Pihlajavesi

A product design contains a large number of decisions about how the product should be made and how it should perform. Dimensions have tolerances, components and materials are selected, and performance requirements are defined based on what we know during development.

Then we manufacture the product, sometimes millions of times, and begin accumulating evidence about how those decisions actually work in production.

We learn which tolerances are easy to maintain and which ones regularly push against the capability of the process. We see how different machines, tools and materials affect variation. Components that looked equivalent during development may behave differently once they are used across thousands of production cycles. Some characteristics require more inspection or adjustment than expected, and some combinations of tolerances create problems that weren't obvious when the individual parts were designed.

I tend to think of the relationship this way: Design determines what should happen. Manufacturing reveals what actually happens.

There is a great deal of useful information in the distance between the two.

What we learn after a design reaches production

Consider a dimensional tolerance on a drawing.

The designer has defined an acceptable range based on the requirements of the product. Once production begins, we can see where the process naturally centers within that range, how much variation it produces and how stable the process remains over time.

After enough production, we may have this information across different machines, tools, cavities, material lots and environmental conditions. We may also know which conditions tend to move the characteristic and which changes have historically created problems.

If the process runs comfortably within the specification year after year, we know that too.

At the other extreme, perhaps maintaining the tolerance requires frequent adjustment. There may be additional inspection, sorting or rework. Certain machines may struggle with it. A tool may need to be replaced more frequently, or normal material variation may occasionally push the process close to the specification limit.

None of this determines what the tolerance should be. The product requirements still have to drive that decision, but the next engineer considering a similar requirement now has real manufacturing history available.

That can become especially important with tight tolerances.

A tighter tolerance may require a more capable process, different equipment, additional inspection or tighter control of materials and process conditions. Those costs may be entirely justified by the function of the product. Production history can help us understand what the requirement has actually meant in manufacturing and whether the expected product performance depends on it.

Process capability is part of that history.

If a new design will use an existing manufacturing process, we may already know what that process has been able to achieve over thousands or millions of parts. We can look at its actual variation and stability rather than treating manufacturing capability as an assumption made during development.

If the new design requires more than the process has historically demonstrated, the engineers can see that early and decide how to address it.

Components develop a history too

The same thing happens with purchased components.

A component may be selected because it meets the electrical, mechanical, performance, cost and availability requirements of a new product. Once it enters production, we begin learning things about it that weren't necessarily available on the datasheet.

Maybe it performs consistently across several products and suppliers. Maybe test results show more variation than expected. Parts from two suppliers may meet the same specification but behave somewhat differently during manufacturing. A particular component may appear repeatedly in investigations involving certain process or test conditions.

After the component has been used for several years, the company may have a considerable amount of experience with it.

The question is whether the engineer selecting a component for the next product can get to that experience.

The information may be spread across production data, automated test results, supplier quality records, process investigations and corrective actions. The people working on the current product may know the history very well, while a design team elsewhere in the company may have no reason to know that history exists.

That is how manufacturing organizations can end up learning the same lesson more than once.

Production problems can tell us something about the design

Manufacturing problems also give us information about the interaction between a design and the process used to make it.

A recurring defect may show that a product is particularly sensitive to normal process variation. An assembly problem may involve the interaction between two tolerances even though both individual parts meet specification. A test failure may eventually be connected to a component characteristic that wasn't significant during development.

Sometimes the issue is more subtle. Production can meet the requirement, but doing so requires repeated adjustment, inspection or sorting.

Those problems are usually investigated because the immediate production issue needs to be solved. An engineer finds the cause, an action is taken and manufacturing moves on.

There may be something in that investigation that would be useful much later, when another engineer is making a design decision involving the same component, characteristic or process.

The connection is easy to lose because the original investigation belongs to a production problem, while the next decision happens in product development.

Cost provides another part of the history

Some design decisions leave a long manufacturing trail without ever creating a major quality event.

A characteristic may require additional inspection on every unit. A difficult assembly condition may create a small amount of rework on every production run. A component may periodically require sorting. Maintaining a particular tolerance may mean more frequent adjustments or more expensive equipment.

Each individual event may be manageable. Across the life of a high-volume product, the accumulated Cost of Poor Quality can be significant.

That history can be useful when a similar decision comes up again.

An engineer considering a tolerance doesn't necessarily need to read years of production records. It may be enough to know that similar characteristics have historically required additional inspection or have been associated with poor process capability.

If that raises a question, the engineer can go further into the manufacturing history and understand why.

The same applies to components. If a part being considered for a new design has been associated with repeated supplier issues, unusual test behavior or production problems in several existing products, that is worth knowing before it is designed into another one.

Making manufacturing history available during design

Much of the information needed to do this already exists.

Products are connected to characteristics and specifications. Those characteristics are produced by processes and have measurement histories. Processes have equipment, parameters and capability information. Components are connected to suppliers and production lots. Problems lead to investigations, actions and results.

Over years of manufacturing, those relationships describe what happened after the original design decisions were put into production.

That is part of Manufacturing Intelligence.

If the relationships are maintained, an engineer working on a new product could ask questions that would otherwise require a considerable amount of searching and experience to answer.

Have we used this component before, and what happened in production?

What capability have we actually achieved for this type of characteristic?

Have similar tolerances been associated with recurring manufacturing problems?

Where are we spending money on inspection, scrap or rework because of requirements similar to the ones being considered here?

Were there previous investigations that would be relevant to this design?

AI makes those questions easier to ask because the engineer doesn't necessarily need to know which manufacturing system contains the answer.

An AI agent could follow the component into the products where it has already been used, find the relevant production and quality history and bring forward investigations that may be worth reviewing. It could look at similar characteristics and show the process capability that has actually been achieved across different equipment or production conditions.

The context around those answers is important. Capability from a different machine, material or process may have little relevance to the new design. A component appearing in several quality investigations does not establish that the component caused the problems. The engineer still needs enough information to understand why a piece of manufacturing history is relevant before using it in a design decision.

This is where years of accumulated manufacturing experience become particularly valuable.

Manufacturers already spend a great deal of effort understanding their processes. We measure what happens, study variation, investigate problems, make changes and learn what works. Most of that work is done to improve the product being manufactured today.

When that history remains connected to the products, components, characteristics and specifications involved, some of what we learn today can also be available to the engineer designing the next product.

Over time, manufacturing gives us a very detailed record of how design decisions behave in the real world. We should be able to use that record when we make the next ones.