This week marked the beginning of my digital manufacturing unit, exploring how a hypothetical corporation could produce 50,000 modular mobile devices per year. The challenge is to design and optimise the lower half of a mobile phone casing using advanced CAD (computer-aided design) tools and manufacturing simulations.

The task is seemingly straightforward but is deceptively complex: 3D-model a precise part based on a construction sheet of measurements, validate the design using a small-scale manufacturing simulation, and then optimise virtual production using specialised software. This process aims to ensure both precision and scalability, crucial for successful high-volume manufacturing.

I started with the CAD modelling process in Autodesk Fusion (Autodesk, 2025), using the detailed construction sheets provided. My focus was on accuracy and replicating the required design with utmost precision. The result is a first draft of the bottom casing, which I will 3D print for next week to establish a baseline for subsequent testing. This baseline will help identify if any dimensional adjustments are required before refining the printing process, and later to detailed manufacturing simulations.

I've attached a short timelapse of the design process, as well as the original drawings to this entry. It took me approximately 60 minutes to complete the design, using the dimensions and a short overview video tutorial.

This week, I 3D-printed the first draft of the bottom casing using my Prusa MK4S FDM printer. For this print, I used grey PLA (Polylactic Acid, a common thermoplastic) at a 0.15mm layer height. FDM (Fused Deposition Modelling) is an additive manufacturing process where layers of melted filament are deposited one at a time to build the object (Acierno & Patti, 2023). The print took approximately 90 minutes to complete.

The prototype is functional but has some cosmetic issues. There is residue on the underside where the part contacted the build plate, likely leftover from a previous print job. This can be fixed by cleaning the bed with isopropyl alchohol, but it does not impact the part’s functionality.

A more significant issue is a small rim of excess material outside the part’s intended geometry (see labelled image). This wasn’t present in the CAD model or the slicer preview (see comparison images). Initially I suspected a layer shift - a problem usually caused by a hardware alignment malfunction; but the rim appeared in the same spot on a second print, after I had verified that the printer was calibrated correctly. On closer inspection, it appears the issue is related to the model’s geometry. The part becomes significantly thinner at this point, which seems to cause a slight over-extrusion due to a stark differential in layer time, leading to the unwanted rim. This is a known issue in FDM 3D Printing often called the “Benchy Hull Line” problem (Prusa Research, 2024).

Despite the cosmetic defects, the prototype passed through the Siemens Festo simulated manufacturing line on the first attempt. It met all critical dimensional constraints and successfully navigated the calibrated processes. My part was the first to be tested, so I was pleasantly surprised by this first-time pass. I attribute this to the accuracy of my design and the impressive precision of the Prusa MK4S, which I’ll continue to use for all prints unless stated otherwise.

The supervisors were impressed with the initial outcome and advised me to focus on improving the cosmetic finish of the part, as this will be part of the evaluation criteria. Additionally, they recommended optimising print time and parameters for future iterations.

Next week, I’ll adjust some settings to address the cosmetic issues, starting with minimum layer time and layer height, and then trying a print with external walls first. To save time and material, I’ll only print the section experiencing this issue, following a rapid prototyping workflow.

This week, I wanted to tackle the “Benchy Hull Line” issue I discovered on my prototype head-on. There’s no universal fix for this problem, as it primarily arises from the thermodynamics of FDM 3D printing. The best way to minimise the issue is by altering the model’s geometry to mitigate sudden changes in layer time. However, since my part has tightly controlled dimensions, that is not an option. Instead, slicer-based optimisations are my primary tool for reducing the prominence of the hull line.

To investigate potential fixes, I tested several slicer settings:

• Control: The baseline print for comparison, this is the first prototype which passed the simulated manufacturing line.

• Added custom supports, 40% infill generation, and external perimeters first: Exacerbated the hull line. 

• Five perimeters: Exacerbated the hull line.

• One perimeter: No observable change from control; hull line persisted.

• Cut section test: No hull line. This was likely due to the different thermodynamic stresses across a smaller print, but this is not replicable for the full part. This confirmed the issue was due to the exact geometry of the part.

• 250% increase in bridging and top solid infill speeds: Slightly worse surface finish, hull line still persists.

None of the changes eliminated the hull line. This strengthens my suspicion that, since the issue is fundamentally tied to the physics of FDM printing this geometry, I would not be able to completely remove the artefact using slicer settings. 

Alongside these tests, I also experimented with disabling "Ensure Vertical Shell Thickness." While this had no effect on the hull line, it did reduce print time by approximately 20 minutes. However, with only two perimeters, the infill became slightly visible through the walls, which compromised the part’s aesthetics.

n.b. all parts were printed using the same spool of grey matte PLA filament, to ensure that material differences did not impact the validity of my testing. 

While working on this, we also had a guided session on PFMEA (Process Failure Mode and Effects Analysis). This tool is crucial for designing robust production processes, as it helps identify and prioritise potential failure modes. I’ll need to apply PFMEA principles later in the project to optimise my production workflow.

One core tenet of PFMEA is the development of Risk Priority Numbers (RPN), a system for scoring in PFMEA to evaluate potential risks in a manufacturing process. It considers three factors: severity, occurrence, and detectability. In this case, the hull line is easily detectable and has minimal functional impact, so it has a low RPN value. A more pressing issue, such as dimensional inaccuracy leading to failure to progress through the Festo Lab would score much more highly, and therefore be allocated a higher severity and be addressed sooner.

For now, I must acknowledge that the hull line cannot be fully resolved through slicer settings. Next week, I will assess an alternative printing orientation, which would eliminate the hull line at the expense of other areas. I will compare and contrast results to determine which path to pursue.

After slicer adjustments failed to eliminate the hull line issue, I explored an alternative approach: changing the print orientation. Printing the part at a 45-degree angle on the chamfered edge introduced both advantages and trade-offs. On the positive side, the hull line is removed entirely, the overall surface finish is comparable and there is less visible scarring from support material. However, this method requires taller support structures, and each part takes approximately 25% longer to print. This impact on overall print time and turnaround needs close assessment, as efficiency is a crucial factor when scaling to our hypothetical production run of 50,000 units.

To evaluate this approach, I tested two different 45-degree orientations. In the first test, I positioned the camera cutout higher from the build plate. Interestingly, there was an unexpected artefact line that appeared approximately 15mm up the print, despite not being visible in the .gcode. Reslicing and reuploading the file resulted in the same issue, ruling out a software error. Additionally, the layer time proved it was not a resurgence of the hull line. Upon close inspection of the resultant part, I identified the issue as stemming from unstable forces acting on the print, likely due to the kinematics of the Cartesian motion system (İncekar, Kaygisiz, and Babur, 2020). Above the artefact line, the print exhibited a significant bowing. This is an issue that would likely not occur on a CoreXY-based system due to its more stable build platform.

In the second test, I flipped the part so that the camera cutout was positioned lower. The results were significantly better - the artefact line disappeared entirely, and dimensional accuracy was restored due to the more stable orientation of the part. There was some very minor support scarring on the undersides of the camera and rear cutouts, but it was much less noticeable than in previous prototypes. This version, printed at 0.2mm layer height, is now the strongest candidate, provided it meets dimensional accuracy requirements in Siemens Festo testing line. 

This new orientation does require support material to print successfully, unlike the previous passing prototype. It uses approximately 1.74 grams of support, compared to the 30.45 grams used for the part itself; whilst this may seem insignificant, it must be considered when evaluating large-scale production quantities. Furthermore, the increase in print time from 60 to 80 minutes for this new part, assuming equal 0.2mm layer height, is also significant; however this may be partially mitigated by the increase in parts per batch this new angle allows for, as now eight units will fit on the bed of the Prusa MK4S, compared to only six at the previous flat orientation, as shown. A reduction in turnaround time could prove especially beneficial if print jobs are being completed a time when nobody is present to clear and restart production, such as overnight, which could lead to an overall improvement in efficiency.

The six piece bed would take 332 minutes and use approx. 172g of material, whereas the eight piece bed would take 625 minutes and use approx. 260g of material. 

The next step is therefore to validate this new orientation against dimensional requirements and continue optimising print time without sacrificing quality.

Additionally, this week there was another guided session on basic manufacturing system design (MSD), which gave a great overview of the basic principles, which integrates well with next week's topics. Good understanding of MSD is paramount for integrating with industry standard practices - which is essential for bringing a product to market effectively using serious manufacturing capabilities. 

The revised part passed through the Festo line flawlessly. I was somewhat concerned that the "staircasing" on the inner guidance lip for the lid may have caused some alignment issue, but this did not come to pass. The fit was good and secure after being exposed to the oven as part of the assembly line.

I have therefore decided to continue with this orientation for the part, and will focus further efforts on optimising print times whilst maintaining part quality, to maximise production efficiency. This part will serve as the benchmark against which any further efforts will be evaluated. 

The primary focus of this week, however, was an introduction to the Tecnomatix Plant Simulation software package (Siemens 2025). This is an extremely powerful platform for production line visualisation and crucially, optimisation - enabling companies to quickly determine more effective ways to manage and arrange their manufacturing facilities to minimise downtime and maximise efficacy. 

I found the software to have a somewhat unintuitive interface, but after an hour or so of guided tutorials, was able to grasp the fundamentals perfectly well. I therefore decided to test my understanding by constructing a rough overview of the Festo simulated production line. I populated each station with the approximate process times, which I recorded with colleagues and calculated last week.

The result of this somewhat crude first run was a calculated output of 109 completed units/hour of operation. However, this assumes an infinite supply of completed lower halves - which is the 3D-Printed part I have been prototyping. 

Over the next few weeks, I will work on refining this simulation by introducing other variables for consideration; including the production times of the printed parts - guided by my prototyping work and further slicer optimisations to reduce printing time. 

This week, we continued learning the plant simulation software.

Building from my basic model of the Festo line, I began looking for points of optimisation. Using the "BottleneckAnalyzer" function revealed that the picking station was the cause of production slowdown; this station is the only one of the six which requires interaction and operation by a human. Improved training or hiring of more skilled workers could lead to an increase, but there are hard limits to the speed a human can realistically work at. Therefore, in a proper production scenario, a manufacturer would most likely either hire multiple humans, possibly in a lower-income economy to reduce salary costs, or seek to automate the picking station if offshoring is not possible or is undesirable. 

Furthermore, my initial model assumed an infinite supply of printed parts. After further work on buildplate layouts, I concluded that nine parts is the maximum that can be reasonably fit in the Prusa MK4S' build volume. Using PrusaSlicer's output, we know that this job will take 12 hours and 33 minutes and use 296 grams of material, using Prusa Research's "0.20mm SPEED" preset. If the printer runs 24/7 at 100% efficiency, it can produce 6,205 parts per year. The calculation for this is as follows:

• Print time per build plate=12+33/60​=12.55 hours

• Build plates per day=24/12.55=1.912

• Parts per day=1.912×9=17.21≈17

• Parts per year=17×365=6,205

• Final calculation: 6,205 parts per year, per printer

Given the scenario calls for 50,000 phone cases, we can therefore extrapolate that 8.06 printers will be required (50,000/6,205=8.06). To account for downtime, print failures and other issues we should round this up to 9 printers. A tenth printer may be prudent if the budget allows, to allow for some breathing room in production quotas, and to allow for some flexibility around machine turnaround, as we did not consider the approximately 2-3 minute window for a human (or robotic arm) to remove the buildplate, install a fresh one and start the next job. 

As with all processes, this would benefit from experimentation. For example, the profiles provided by Prusa Research are robust, and produce quality-looking and strong parts, but are somewhat conservative. We may be able to produce parts of a similar quality using a larger layer height and/or higher speeds, which could significantly reduce the time per buildplate - however, for a first production run it seems prudent to make use of the tried-and-tested presets provided by Prusa Research for their machines. If production is running smoothly, we could theoretically allocate one of the ten printers to test a further optimised profile, and evaluate the results. 

Taking a break from the production line for this week, the focus was on the standard industrial approaches for machine and process evaluation. When deciding to invest in new machinery, businesses must carefully determine whether the purchase will be suitable for their production goals, both financially and in terms of accuracy. 

One way of doing this is through evaluation of process capability, using Cp and Cpk. Cp measures whether the process spread is narrower than specification width, whereas Cpk measure both the centricity of the process as well as the proces spread relative to specification width. These can also be expressed with equations as follows: 

• Cp = Allowable tolerance / 6*σ 

• Cpk = min{ (USL- mean) / 3*σ , (mean - LSL) / 3*σ }

• Where: σ = Standard Deviation & USL/LSL = Upper/Lower Specification Limit.

The overall goal of this system is to ensure that a potential machine is capable of producing parts to the required specification, which is paramount if a quality product is expected. 

These methods require sample parts to be measured, and the deviation calculated from the expected dimensions - no process is perfect and there will always be some deviation, however minor. However, there does need to be a practical acknowledgement of the acceptable limits both of the parts and of the measuring methodology, as it too will have various nuances affecting reliability of results (Fanton, 2019).

Nonetheless, I thought it prudent to conduct a rough estimation of the accuracy the machine I am using for this project - the Prusa MK4S. Whilst the manufacturer does not provide a full suite of test results, they do share an image on their website of measured dimensional accuracy of a part using a CT scanner - from this we can extrapolate a claimed dimensional accuracy of ~0.3mm deviation (Prusa Research, 2024).

For the evaluation of my personal machine, I have printed ten 20mm³ cubes, and measured all of them using a vernier gauge, ensuring they were all in the same orientation. I then calculated a mean dimensionality to determine if my machine meets Prusa's stated values. The results were as follows:

Measured Widths (mm):

• 19.91, 19.92, 19.93, 19.96, 19.98, 19.985, 20.00, 20.01, 20.04, 20.05, 20.05

• Mean: 19.985 mm

• Range: 19.91 mm – 20.05 mm

• Spread: ±0.07 mm from the mean

Therefore, I can conclude that my machine is operating well within parameters. This is especially impressive, given I personally assembled my printer. As thermoplastics undergo changes in dimensionality through heating and cooling cycles, the slicer is automatically compensating for this by slightly adjusting the dimension of parts, so that they achieve the desired measurements once cooled. This is a property that is user-adjustable, but this simple test demonstrates that the default value is well suited for this specific PLA.

This therefore confirms that the MK4S is more than accurate to produce the requiste 50,000 phone cases at a good level of accuracy; whilst a definite threshold was not provided, a tolerance of >0.1mm is sure to be more than sufficient. 

Fanton, J.-P. (2019) ‘A brief history of metrology: past, present, and future’, International Journal of Metrology and Quality Engineering, 10, p. 5. Available at: https://doi.org/10.1051/ijmqe/2019005

Original Prusa MK4S (2024) Available at: https://www.prusa3d.com/product/original-prusa-mk4s-3d-printer-5/ (Accessed: 8 April 2025).

Next up, the Industrial Internet of Things (IIoT). The Internet of Things paradigm describes interconnected systems of physical devices, from small parts such as sensors and switches to entire machines and processing units. These devices can all communicate with each other over a network, which could be a local area network, or the global internet (Kopetz and Steiner, 2022).

IIoT is the industrial application of this idea, utilising an IoT system with upgraded security, robustness and analytical capabilities. IIoT solutions are often integrated with cloud services, enabling secure device communication across different sites in near real-time, as well as enabling constant data backup (Li et al., 2023). These systems are typically managed via an intermediary platform or dashboard, which gives users control over all connected devices from a central interface. This week, we were introduced to one such platform: Siemens Insights Hub. 

Insights Hub allows us to monitor, control and analyse the CP Lab remotely, made possible by a cloud gateway—a processing unit connected via Cat6 cabling to each of the CP Lab machines. Once configured, a system like this could enable so-called "lights out" manufacturing; where the entire production line can run uninterrupted without direct human supervision. This could unlock massive efficiency gains, allowing for overnight or weekend production and potentially increasing yields by up to 20%. 

We also briefly covered the initial setup and operation of a Node-RED, web-based interface for a 3D-Printer. Node-RED is an open-source programming solution, which could be configured to control a variety of printers from different manufacturers if required. However, if only a single brand of printer is being used, it would likely be an overly complex solution as many manufacturers now provide their own cloud-based interfacing dashboards. Nonetheless, it is 

Kopetz, H. and Steiner, W. (2022) ‘Internet of things’, in Real-Time systems. Available at: https://doi.org/10.1007/978-3-031-11992-7_13

Li, F., Lin, J., & Han, H. (2023). FSL: Federated sequential learning-based cyberattack detection for Industrial Internet of Things. Industrial Artificial Intelligence, 1(1). https://doi.org/10.1007/s44244-023-00006-2

According to the Digital Twin Consortium, a digital twin is a "virtual representation of real-world entities and processes, synchronised at a specified frequency and fidelity."

Last week, some of the benefits of the IIoT were discussed, but the true value of such systems lies in digital twins (DT). DTs allow us to monitor not just individual components, but entire systems and integrated processes; this in turn allows for a massive amount of data to be collected, then aggregated and processed. This enables a wide variety of concepts, from performance flow optimisations and real-time fault monitoring, to predictive maintenance - where machines can be maintained preventatively before a fault is likely to occur, significantly reducing downtime and repair costs.

A similar approach can be applied  at a factory-wide level, analysing the entire production line to find opportunities for optimisation to a degree that would never have made economic sense to pursue in the past. This process is accelerating further with the growing advancement of artificial intelligence algorithms, able to decipher large data packets and recommend alterations faster than human analysts ever could.

We were provided with hand-on training with DTs, using another facet of Siemens Insights Hub, called Industrial Edge. This package presents near-real time data from the Festo Lab production line, allowing for quick turnaround data analysis. However, as a proprietary Siemens service, it is only compatible with Siemens hardware and their approved vendors - a similar practice of closed-source technology is present throughout industry; a more global, unified protocol could allow for even greater insights if service providers were to collaborate. 

Once we had the digital twins set up, and analytics running smoothly, we could further take advantage of the interconnected capabilities afforded by using the data from Siemens Insight hub directly inside the Plant Simulation software via a software plugin. This allowed us to directly simulate the production line, using the real data from the real machine to extrapolate production scenarios. Specifically, we were able to simulate how long it would take the line to complete the 50,000 required phone cases. If it ran non-stop, this would take 18 days and 12.5 hours. However, this is not yet directly applicable to the real world, as it does not account for the time required to produce the phone cases - a process which takes far more time than the assembly. 

As, in this scenario, we are only contracted to produce the phone cases and then assemble the final product, we can work out an optimal production schedule based on our production time estimates. A single printer (Prusa MK4S) can reliably produce 17 parts per day. With a small farm of 10 printers running consistently, this scales to 170 parts. Multiplied over a 7 day week, this is 1,190 parts per week. Within a 52 week year, which is the length of our contract, our production output would be 61,880 units - over 20% more than required, leaving some breathing room to account for machine repairs, scheduling inconsistencies and print failures. 

Running the plant simulation again, we can project that it would take 10 hours and 40 minutes to assemble 1197 cases; when accounting for the additional 20 minutes required for a statutory break period (Government Digital Service, 2012), means we could process 7 days of production output in one 11 hour shift. This would allow us synchronise production and operate a just-in-time manufacturing model. Instead of stockpiling parts, we could run one long assembly shift per week. This would reduce the need for storage and more importantly, maintain a tight feedback loop between production and quality assurance (QA).

As the Festo lab line acts as both assembly and QA, this regular cadence would help to catch any printing issues early, rather than discovering them weeks or months later. A calibration print could be run at the start of each assembly shift to validate the accuracy of the line. The simulation also revealed that the Pick-by-Light station is the current bottleneck in the Festo line, with an average operation time of approx. 32 seconds. - though this was recorded from an untrained operator, so it's likely that this could improve with experience. A potential optimisation could be to split the shift between two trained operators on shorter shifts. This approach would reduce fatigue and add redundancy in case of absence or training needs. 

In conclusion for this week, based on our current data, both part production and final assembly will be scalable with the right planning, and there are already clear opportunities to optimise both time and resources. 

Government Digital Service (2012) Rest breaks at work. Available at: https://www.gov.uk/rest-breaks-work (Accessed: 1 May 2025).

The close integration of additive manufacturing and digital manufacturing simulation has been eye-opening. Utilising data from my personal 3D Printer, I was able to model an output of 17 parts per day per machine, demonstrating the feasibility of scaling production. This approach allowed for rapid prototyping and continuous iteration throughout production.

Using this data—especially real-time insights from the digital twins of the Festo manufacturing line—I was able to simulate the entire production process in Plant Simulation. By trialling the processes in this way, I was able to identify bottlenecks, test potential changes and accurately predict the line’s performance across the full 50,000-part contract by physically printing just a handful of prototypes.

I was taught and then made use of PFMEA, MSD, and metrology principles to justify regular calibration runs runs before each proposed day of operation for the Festo line. This will ensure ongoing accuracy and supports my proposed just-in-time manufacturing model, ensuring that flexibility of the process is maintained.

I went through a rapid prototyping phase where, over the course of a couple of weeks, I refined the printing process of the geometry to minimise machine turnaround time while eliminating unsightly defects, such as the hull line artefact. The number of parts on each printer scales in a linear fashion, and with my proposed number of machines supports a feasible production schedule with minimal staffing costs. This level of planning was only made possible by the insights from the use of the digital twins and simulations.

Through this development log, I have come to appreciate the value of a rapid-prototyping workflow, especially when integrated with the detailed level of predictive modelling offered by plant simulation. I personally feel that the design of the phone case is dated and could be optimised - it is overly bulky and uses more material than is necessary. Nonetheless, the task was to refine the manufacturing process for that specific part. I am pleased with my efforts and believe I have proposed a realistic and efficient framework. 

If I were to produce this part again, I would spend further time exploring printing optimisations, as I feel more progress is possible in this area. Similarly, with the increasing market position of faster CoreXY motion systems over the more standard Cartesian setup present on the Prusa's, using different printer(s) for the farm could save a significant amount of time. Finally, I would explore integrating a more robust form of traceabililty to each part, ideally using some sort of QR-style code printed as part of the unseen inside of the case; this would improve tracking of parts and lead to faster issue resolution, as each part could be tied to a machine and time of production.