Elementary, My Dear Watson — Mechanistic Models (Cost Modeling Methodology Part 3 of 3)

This weekend, I'd like to welcome Eric Hiller back to Spend Matters. Eric, who is founder and Chief Product Officer at aPriori, will be contributing a series of guest weekend posts on the subject of driving Spend Management upstream. And this is the eighth installment. Please join in welcoming Eric to Spend Matters!

Sherlock Holmes was a character famous for being able not only to provide the correct answer to a mystery, but also explain the exact reasoning behind the conclusion and all of the cause and effect relationships. Mr. Holmes could tell you not only what, but also why. Mechanistic models are much more similar to Sherlock than the Oracle of Delphi (Empirical Models). A mechanistic costing methodology breaks cost down to its individual pieces. It is a bottoms-up approach that starts with the basic principals of design, sourcing, manufacturing, and finance.

Although it would be possible to make a mechanistic component-based model (models tied to what the type of component being manufactured is), mechanistic models are much more likely to be process-based (models tied to how the component is manufactured). Mechanistic models start with inputs like the geometry of the part or product, the sourcing strategy, production volumes, material, the process routing, machine selections, etc. Many of these inputs may be the same ones that the empirical model starts with. However, instead of a top down model based on statistics, the mechanistic methodology seeks to map the inputs to the processes in a clear, understandable, and logical way. Simply put, a mechanistic model simulates the physics of a manufacturing process to produce the physical outputs of time and material utilization and then overlays the costing method of a given factory or supplier to convert that time and that utilization into the universal language of the business: cost.

A short example of a Mechanistic Cost Model
For example, let’s say that we are modeling the manufacturing processes used to make a machined metal part for example, milling, drilling, turning, etc.). In a mechanistic approach, we would first determine what Geometric Cost Drivers™ are pertinent to these operations. For example, we might determine which of our factories’ standard stock sizes will produce parts with the least waste (highest utilization). Based on the number of geometric artifacts (holes, surfaced faces, pockets, etc.) and their direction, we can determine what direction will provide us the most effective first machining fixturing set-up, as well as if other set-ups will be needed. From the size and number of the various pockets in a given set-up as well as the part material, we select a tool from the machines’ library that will rough out the part. We can also look at the surface roughness of various faces to determine how many passes we will need to make with our tools and calculate the time for this operation. Based on the diameter and depth of each hole, we can calculate how much material must be removed and how long this will take. Based on the number and size of each round hole we will know how many times we will have to change drills. We could continue, but this is enough for a general understanding of how a mechanistic model works. In this example, some of the Geometric Cost Drivers™ include volume of material removed, surface roughness, number of round holes, hole diameter, hole depth, etc. There are other cost drivers, both geometric (mass and size of stock) and factory specific (time standards) that will drive how long it takes to load the blank on the machine and unload the part. Sourcing specific cost drivers (material supplier and prices) and Geometric Cost Drivers(tm); (size of stock) will determine how much material is used, the utilization, and the material cost. Sourcing information like labor rates and overhead rates translate physical quantities like time, utilization, mass, etc. into the financial domain of dollars.

Is it too complex?
Although, the mechanistic approach is based on the physics of the process, it does not have to be unreasonably complex. Sometimes people ask how it can be possible to model the very complex world of manufacturing without getting mired down into an academic science project. The solution to this question is the fact that many process considerations necessary for manufacturing a part may be simplified and abstracted in the cost model for a process. For example, while it is true the design of sprues and runners will certainly affect the cycle time of the injection molding process and the yield, we can assume that competent people are designing the process and tooling and will set up the sprues and runners so the yield will be normal for the process.

For options that cannot be abstracted or assumed, a properly constructed Enterprise Cost Management (ECM) platform will work in conjunction with the mechanistic model to complement it. In these cases, the ECM platform will have a process set-up screen with process options that can be specified by a knowledgeable user, but are selected for other users based on Geometric Cost Drivers(tm) or by defaults. By abstracting the process attributes and leveraging the power for an ECM platform, mechanistic methodology for process modeling becomes very practical.

The disadvantage of a mechanistic approach is that it can take longer initially to understand the manufacturing process at a fundamental level and map all of the geometric and non-geometric cost drivers to the process model. Even with reasonable abstraction, the original process models must be built with careful consideration and understanding of how the processes work. An empirical approach frees the modeler of needing to understand what drives time and cost at a deep level, because the power of the statistics allows the modeler to ignore how the process really works. Therefore, the construction of mechanistic models requires expert cost and process modelers with extensive knowledge of manufacturing, sourcing, and design.

However, once the mechanistic model is finished it has many advantages over other approaches.

• Speed -- Like empirical models, mechanistic models can provide very fast results. There is no need to wait for an expert to provide an opinion or a quote to return. As fast as inputs are provided, the model can calculate results.
• Consistency -- Unlike Quoting and Expert opinion, a mechanistic approach is very consistent and able to provide Relative accuracy between designs, suppliers, etc.
• Stable & Bounded -- Typically mechanistic models are bounded and do not produce ridiculous results, even when used beyond the regime of data against which the model was originally tested. Furthermore, mechanistic cost models tend to be naturally bounded, because they are based on physical machines, suppliers, etc. that have size, capacity, etc. limits that can be easily built into the model to warn users when input to the model seems out of the ordinary.
• Clarity -- Because the model is built bottoms-up, it is typically very clear what cost drivers and logic in the model are responsible for changes to time and cost. Therefore, it much easier to pinpoint the cause of any problems with the model, improve it, and extend the model to cover new situations, processes, and technology.
• Learning -- The clarity a mechanistic model provides an opportunity for users to clearly understand what drives cost and learn how to source, design, and plan in ways that minimize cost. This allows a co
mpany to capture and institutionalize tribal knowledge and extend it throughout the organization and the development cycle, resulting in the elimination and/or avoidance of cost.
• Extensibility -- The greatest advantage of a mechanistic model is its extensibility. The Mechanistic methodology breaks the model into a logic layer, business rules, inputs, and a data layer. The logic layer is based on the physical reality of the process modeled, but the data may be based on the machine, the facility, etc. For example, there are many different types of injection molding machines, but within the bounds of a few machine options, the physics of injection, cooling, and ejection is very similar for all machines. Therefore, one generic physics-based mechanistic model may be constructed for all machines. Every supplier or internal factory may have a different set of machines, but to generate an accurate cost only requires the data that feeds the model to change, not the underlying logic. The advantage of this extensibility is hard to overemphasize. It allows the same process model to be applied to any factory (internal or supplier) that uses this process. Including the new factory in the model requires only the gathering of the data for the factory, including which machines are present, which routings are allowable, their business rules, the physical speeds, feeds and limitations of the machines, and the cost accounting rates of the factory. The extensibility is also useful for capital planning for factories, production lines, and machines that do not yet exist. For example, if the machine parameters for a new machine the company is considering buying are added by creating a new virtual machine in the factory’s inventory, the company can determine how long it will take to break even on the potential investment.

A Summary Comparison of Cost Model Methodologies
In the last three post, we have cover four different methodologies for Product Cost Models. The links to the post are here:

The table below lets you compare all of them at once.

Author: Eric Arno Hiller
Founder & Chief Product Officer of aPriori

Share on Procurious

This site uses Akismet to reduce spam. Learn how your comment data is processed.