Engineering Capacity Planning
1 Introduction
Stadler relies primarily on experience and intuition when making capacity decisions. Although this strategy has proved historically successful for the company, there isn’t a concise feedback loop to validate and improve the predictions. It is difficult to communicate expectations across the project phases with regard to predetermined budgets. Managers with limited understanding of the overall workflow are at risk of understaffing their departments at critical phases. This in turn leads to missed payment milestones, and management is pressured to add emergency personnel to make up for lost time.
Although coined for a separate industry, Brooks’ Law takes shape, and that is “adding manpower to a late software project makes it later”. Brooks’ Law describes a scenario where total productivity declines due to resources being diverted to onboard potential contributors.
Broadly speaking, strained capacity
increases the total cost by means of delay costs, as visualized in the figure
below.
The objective of this document is to
quantify the capacity cost and thereby reduce delay costs with regard to
engineering.
2 Quantifying
Outputs
The deliverables of train engineering are defined in the project vehicle specifications. Each system is classified according to the DIN standard and evaluated per chapter in the Technical Compliance Specification Matrix.
For the purposes of this document,
we will be focusing primarily on the delivery of engineering drawings for use
by Procurement and Production.
2.1 Modeling
and Drafting
Drawings and Bills of Material are arranged in the train structure by configuration identifiers known internally as “articles”. The articles are paired by child/parent sequences that reflect the design levels of the project. This structure can be exported from the Project Data Management Software.
These exports contain between 40,000 and 100,000 articles, however, articles used across multiple configurations appear as multiple rows. When duplicates are filtered out, projects contain less than 10,000 unique articles.
The most important document produced by engineering in the project structure is the installation drawing. All updates to the vehicle structure are propagated immediately to the installation drawing to ensure that production has the latest Bill of Material and instructions for use in final assembly.
The schedule of installation releases can be tracked across projects to gauge the output of engineering teams. Below is an example generated from an existing vehicle structure.
However, it is worth noting that these releases are the latest revision according to SmarTeam. In other words, the “final” revision of the material. It would be worthwhile to see the release dates for initial revisions, but that data was unavailable as this document was being written.
Figure 2 – Sample Project Installation releases per Project Month
The amount of hours required to produce an installation vary depending on the system, but when analyzed on the lowest level tasks, an bottom-up estimate can be created. A sample estimation is provided in the table below:
Table 1 – Engineering Cost of Project by
Installations
Installations are broken into three
subcomponents: Hardware, Parts, and Assemblies. Hardware is defined as
fastening elements with pre-existing 3d models and articles. They are inserted
directly into installations and assemblies and positioned.
Parts are defined as the lowest
level components of an installation and are fabricated from raw material. Parts
have 3D-models and drawings which must be created by engineering.
Assemblies are defined as a collection of parts or subassemblies positioned together in some configuration. The installation itself is an assembly at a certain level. Assemblies require engineering to create 3d models and drawings to describe how the components fit together.
The blue inputs in Table 1 were
based on opinion from various engineers familiar with design work. The yellow
inputs in Table 1 were conservative estimations of average assembly and part
quantity per installation. Given those inputs, a metro project calling for 174
installation drawings would likely require 55,912 engineering hours dedicated
to the production of 3D-models and drawings.
2.1.1 Statistical
Analysis of Train Structure
The yellow inputs in Table 1 can be calculated by compiling a list of each installation in a project. A script sorted through the lower design levels, designating the lowest level replaceable units as “parts”, and everything in between as “assemblies”. Duplicate articles were counted once and then ignored. This prevents the work generating the article from being counted twice. Standard parts, such as hardware and consumables, and catalogue parts were also excluded from the count.
Given the blue inputs from Table 1, the following table shows the installation creation time for various projects:
Table 2 – Average Quantities of Assemblies
and Parts with 95% Confidence
|
Project |
Type |
Installation Sample Size |
Mean Quantity of Assemblies |
Mean Quantity of Parts |
Mean Hours per Installation* |
|
Project 1 |
Metro |
163 |
23.5 +/- 5.6 |
5.5 +/- 1.5 |
172.9 +/- 32.5 |
|
Project 2 |
EMU |
314 |
18.8 +/- 4.1 |
12.8 +/- 3.2 |
186.1 +/- 33.7 |
|
Project 3 |
DMU |
234 |
31.5 +/- 6.4 |
8.1 +/- 4.0 |
222.7 +/- 47.2 |
*Hours data taken from Table 1
By taking the structure data from
multiple projects per platform, then a baseline can be established according to
the platform, which can then be the assumption for future projects.
3 Quantifying
Inputs
3.1 Simulated
Resource Planning
When a project is in the bid phase, there is a budget estimate provided by engineering. The weights of these stages are determined based on the novelty of the systems. An existing vehicle platform would need less conceptual work because the system specifications are already defined. If the vehicle is being prepared for a different continent, then the weight would lean towards drafting and design.
If a large percentage of existing articles must be new designs, then the cost of that effort can be estimated using similar vehicle structures. Those costs would land primarily in the beginning to mid stages of the project schedule.
The project schedule is split by the following milestones:
PDR
(Preliminary Design Review)
· The first phase that begins upon successful bid. Vehicle specifications should be completed by this stage.
IDR
(Intermediate Design Review)
· The second phase. Drawings should be released and sent to Procurement for advanced pricing and lead time information.
FDR
(Final Design Review)
· The third phase. Drawings should be finalized and FAI’s should be complete.
Trainset
1 Begin Assembly
· This begins engineering changes and production support.
Trainset
1 Delivered
· This marks the end of engineering changes and production support.
Note that assembly of Trainset 1 typically begins prior to closure of the FDR. In some projects, the FDR isn’t closed until Trainset 1 is delivered. However, the resources allocated to production support can be added to the existing FDR resources.
A typical loading curve is shown
below:
Figure 3 – Typical Mechanical and Electrical Team Loading for a Project
Note the how the mechanical loading varies greatly compared to the electrical loading. This is typical and stems from the nature of the work being performed. Mechanical must produce a substantial amount of drawings prior to the IDR, where the electrical team must create relatively few schematics. After the IDR, the required effort to revise the released documents and validate material is about the same between mechanical and electrical.
The loading plan can and should be modified according to updates to the project schedule or department resource availability. The loading plan should always reflect the latest information.
Consider the following example
loading curve, taken midway through the project:
Figure 4 – Updated Loading Curve vs Typical Curve
In month 13, it was announced that
the project would be undergoing significant redesign. A second PDR was
scheduled, other milestones were delayed, and more budget was granted to
engineering. The old loading curve (shown in red diamonds) was no longer
beneficial to planning activities.
3.2 Case
Study
Although the loading curve is by no
means a perfect prediction of the future, it is mandatory in the acquisition of
resources for critical points in the project. Consider the case of relatively small metro project, where the actual loading data is presented in the plot below:
Figure 5 – Actual loading curve with original milestones
This particular project did not have an IDR. The drawings were scheduled to be released by FDR on month 22. It’s immediately noticeable that the peak of the mechanical team loading did not occur prior to the FDR.
The graph tells the story. At month 17, it was became clear that the initial team of 15 engineers would not be able to deliver all drawings on time. The team quickly added more engineers, but it was not enough to reach the original milestone date. Project Management responded by negotiating the FDR to be delayed until the delivery of trainset 1. However, without completed drawings to initiate the supply chain, delivery would inevitably be delayed as well. After the FDR was delayed, more team members were injected into the already late project, where they worked for 7 months.
Due to the late intervention, the completion of trainset 1 was delayed by 7 months. The FDR, which was tied to the completion of trainset 1, was delayed 16 months.
The figure below shows the same loading
curve with the milestones as they actually occurred.
Figure 6 – Actual loading curve with delayed milestones
The begin of Trainset 1 essentially became the de facto IDR, where released drawings were required by procurement rather than the customer. Production support and engineering changes were rolled into the FDR efforts. Overall, the loading curve ended up proportionally the same as the typical loading curve, except with delayed milestones and absorbed production support budget.
Milestones should drive hiring and
capacity planning because it’s ultimately available resources that determine
the actual milestone completions.
4 Data
Driven Decision Making
4.1 Department
Hiring
The ability to hire engineers is
controlled primarily by division management. The annual budget takes future
project loading into account and proposes a headcount for the department.
Figure 7 – Projected project loading simulated two years into future
The simulated loading curves from all current and future projects can be combined into a comprehensive loading curve for the department. Overhead and sales activity are also to be included in the overall loading curve.
The average clocking activity is around 90% of the total headcount. If you multiply the 52 weeks in the year by 40 working hours, you get 2,080. However, after taking PTO and holidays into consideration, the actual working hours per year is 1,889.
It is not warrant for concern when the current plan shrinks the employee count to zero. In project driven companies, the sales department is aware that they need to win projects every 9 months to sustain the engineering department. Dips are common, and available resources can be redirected to other activities.
To propose a headcount using potential projects, there are a few methods available. One simple approach is the “0, 70, 100” method. Projects are given a likelihood of either 0%, 70%, or 100% (whichever is closest, so 30% likelihood would be counted as 0%). The project hours are then multiplied by that factor and added to the current plan. Once all the potential projects are added to the plan in this way, you can set the headcount to 80-85% of peak loading and plan on overtime.
For example, consider a potential project with an estimated 50% chance of winning. If this project were to have a peak loading of 32 engineers, we would use the nearest probability factor of 70% and set the peak loading to 23 engineers. After adding that project to the total, we could see a total peak loading of 50 engineers at some point during the next year. Our proposed headcount would be 40 engineers, or 80% of that peak.
4.1.1 Stochastic
Analysis
The “0, 70, 100” method is a good rule of thumb, but there are more sophisticated methods for developing a hiring strategy, and one such method is Stochastic Analysis.
Stochastic Analysis is a matrix operation that creates a list of possible outcomes with associated likelihoods of occurring. The likelihood for an individual event to occur is a value between 0 and 100%, where the sum of all possible likelihoods equal 100 percent.
The following table demonstrates stochastic analysis of three projects. The inputs for each project are percentage likelihood of successful bid.
Table
3 – Stochastic Analysis using three inputs
Three projects gives us eight
possible outcomes. The likelihood for a project being the only successful bid
are found in the diagonal squares where the project is matched with itself.
These are highlighted cells in the table. The likelihood for two projects being
successful are found where the associated projects overlap. For example, the
likelihood of Project 1 and Project 2 being successful is 6%. As more projects
are simulated in this way, there are exponentially more potential outcomes.
4.1.2 Strategy
and Risk Mitigation
To use the results of the stochastic analysis in the hiring plan, each of the potential outcomes need to be modeled as a loading curve in the projected project loading (see Figure 5).
From there, there are two measurements from which to make hiring decisions. The local maximum can be taken by using curve data within the year timeframe, or Year-End loading. This local maximum is useful information for a proposed hiring budget that only takes the year into consideration, which syncs with the annual budget planning. However, due to the restricted timeframe, Year-End loading gives a potentially false read on project loading that could grow substantially in the following year, which would leave the department understaffed.
The alternative to a Year-End measurement would be to simply take the global maximum of the loading curve, or a Project measurement. The Project measurement is useful for determining if a department is inevitably too large or too small.
Consider the following table, where the maximums of each potential loading curve are factored according to Year-End and Project measurements.
Table
4 – Example Year-End and Project maximums for potential loading curves
The maximums of Project loading will always be equal to or higher than the Year-End maximums. If the peak loading for a project occurs within the same year, then the values will be identical.
The next step is to combine the loading maximums with the stochastic data we generated in Table 3.
Table
5 – Risk Assessment through Stochastic Analysis
The peaks of the various loading possibilities are listed in the “Peak Demand” column. Those values are compared against the other potential peak loadings and their associated likelihoods in Table 3. The risks are split into three categories: perfect loading, over absorption, and under absorption.
Over absorption (OA) means that the department is understaffed to accommodate the project workload. That drastically increases the risk of project delays and therefore total costs as shown in Figure 1.
Under absorption (UA) means that the department is overstaffed according to the scheduled workload. Unless each staff member has an assigned task where they can hasten the closure of a project, then the department might initiate lay-offs.
It is management’s decision as to what the final hiring target will be. Table 5 allows the manager to weigh their decision against the possible outcomes and prepare contingencies to mitigate risks for either delays or lay-offs.
For example, using the information from Table 5, a hiring target between 55-62 would yield the lowest risks of either over absorption and under absorption. If the project has tight milestones, then a more aggressive hiring strategy would be preferred. If a project has loose milestones, then potential delays would have less impact on company cash flow.
Using the “0, 70, 100” method on the same sample data in Tables 3, 4, and 5 would result in a hiring target of 63. Although this is very close to the proposed range of 55-62 generated by stochastic analysis, it has a high risk of under absorption that isn’t measured and communicated to management. Therefore, no substantive contingencies can be made. As a result, it’s no surprise that Stadler faces sudden lay-offs when bids aren’t successful. There isn’t enough time to find productive tasks for individuals hired for an unsuccessful bid as no effective contingencies had been prepared.
5 Conclusion
Timely resource allocation is critical to the success of any project. In order to effectively plan and budget resources, the inputs and outputs must be defined in quantifiable and functional means. Proper planning requires that current assumptions are mapped out according to the predetermined project milestones and updated frequently. There are countless unpredictable factors that may steer a project outside of initial parameters, but the ability to rapidly identify, quantify, and respond to those changes are a pure measure of organizational capability.
In the context of an engineering department, the outputs have been defined as the submission of completed models and drawings. The inputs are the project milestones by which the outputs must be delivered and the vehicle specifications that define the quality of the outputs delivered. Both outputs and inputs are clearly defined and measurable with respect to project hours allocated in the budget.
Through simulating resources across multiple projects, both actual and theoretical, it is then possible to asses risk through stochastic analysis. Management can then predict the impact of necessary resource decisions, and plan contingencies to mitigate negative consequences. Managers can lay the groundwork for future hiring campaigns with the knowledge that their team will likely be understaffed if the company secures more projects. Alternatively, managers can cool hiring activity and begin transitioning team members to other functions if future under absorption is a high probability.
By properly maintaining capacity costs according to project demand, delay costs can be mitigated and catastrophic budget overruns can be avoided.
Comments
Post a Comment