Two main solutions have been developed for Supply Chain in Theory of Constraints:
- DBR: Drum Buffer Rope: In an MTO (Make To Order) environment the desired results cannot be achieved due to interdependency and variability in successive processes. There is a better alternative as presented in The Goal book. You can try it with the simulator and find an add-on DBR software in your ERP software. I wanted to introduce 3 different simulators in this series:
- 1- Dr. Kelvyn Youngman / https://www.dbrmfg.co.nz / excel-based simulator, please refer to related article
- 2- Alex Knight / QFI Consulting / online simulator, see related article
- 3- Dr. Alan Barnard / Goldratt Research Labs / online Dice Game simulator
- Replenishment - Completion: For MTS (Make To Stock) or MTA (Make To Availability production) environments the desired results could not be achieved due to too much / too little stocked SKUs despite the high inventory level. Goldratt suggests a better alternative as presented in Isn't It Obvious? book. Similarly, you can try it with a simulator, and find an add-on Replenishment software in your ERP software. In this series, I will introduce 2 different simulators:
- 4- Matias Birrell Rodriguez / Goldfish / FILLRATE100sim / a simple online simulator, you can refer to the related article
- 5- Dr. Alan Barnard / Goldratt Research Labs / Hannah's Shop simulator based on DDMRP (Demand Driven MRP - Demand Driven MRP) online, see the related article
3- Dr. Alan Barnard- EMGoldratt / Goldratt Research Labs online simulator
It is the original and, in my opinion, the most interactive simulator. The working time is 100 days, not 20 or 30 days, and you can shorten/extend it as you wish. You can edit the average capacity and variability for each operation. You can accumulate any amount of stock between operations and if you want you can start the simulation with full WIP for every operation. The goal and plan will be updated according to your choice, but the demand remains unaffected. You can set the demand as constant or uniform - evenly distributed. You can change the selling price of the product, the raw material cost it contains, the daily overhead, the unit cost of the labor, the target achievement threshold, and the premium % that will be received when the target is met. You can examine the 5 installed scenarios by working comparatively for any period you want.
There may be people who find the variability with the dice excessive compared to real life; but in real life, have you ever made an extraordinary effort for an "urgent" order? Or did you put your work aside in favor of another order and keep it waiting? As a result, the purpose of the tool is not to replicate your real life, but to provide a learning experience by simulation.
We are all used to the normal distribution (bell curve). When you roll 3 dice together, the normal distribution is already obtained. In this model, there are 5 operations, that is, five dice come together. When you run it several times, the output of the production line in the simulator shows a normal distribution.
Our team all worked together on the simulator. It was an interesting learning process and I am sharing our experience step by step in the same order:
The base scenario is the opening setup of the simulator: We have a 5-operation process that has zero stock among them (one piece flow), average capacities fully balanced (all 3.5 units), matching fully to takt time (demand is 3.5), with same variability (1-2-3-4-5-6). There are no real-life disturbances such as absenteeism, unskilled labor, poor quality, poor efficiency, fatigue, shared operations, machine breakdown, missing confirmations or raw material, diversions, returns, overtime, work accident, poor ventilation - lighting - heating, ... Here there is none of them! This is a production environment that is too ideal to be true! Our target is to produce 350 units at the end of 100 days. Since we have 5 stations, our flow time should be close to 5 days. There was no WIP in the beginning, so there will be very little WIP at the end. Surely we will make a profit. Let's go!
The model provides a wide variety of feedback: Efficiency % by process, min-average-max values of planned capacity and actual production, idle-normal-blocked distributions of WIP, production output number, target achievement %, Gannt Chart of processes, flow time min -avg-max, WIP min-avg-max values, financial income statement, profit%, ROI%,...
We ran it 3 times and record the average: We were able to produce 155 while we were expecting 350, we expected 5 days for the flow time, but it was 30 days, and we said that there would be no WIP in the line, but 105 units of WIP were accumulated. We had no profit and an ROI -41%. Why?
Let's give it more time: Sometimes when things don't go well, we say, "Let's be patient, we need some more time, it'll get better". We tried again, extending the 100-day run period to 200 days: we were able to produce 333 while expecting 700, flow time increased to 57 days, WIP increased to 199, loss increased, ROI -43%. So it doesn't get better with just waiting, patience is not a cure.
Let's put 1 unit of stock in between: We abuse the one-piece flow, thinking it will help absorb the variability in the processes. We ran it for 100 days again: we were able to produce 212 while expecting 350 (from 155 to 212), flow time decreased from 30 days to 23 days, WIP fell from 105 units to 77 units. We're at a loss, ROI -36%. Adding extra stock in between processes increases the performance.
Let's make WIP 5 units (the same amount as the number of stations): According to Little's Law, more stock than the number of stations will not work. We were able to produce 271 while expecting 350 (increased from 212 to 271), the flow time decreased from 23 days to 15 days and WIP decreased from 77 to 49. We're at a loss, ROI -2%. Stock works. It takes time to fill the line with stock and there is a risk of missing the deadline. Of course stocking up is not free but it seems to work. Can we increase it a little more?
Let's stock more than the number of stations (8): We were able to produce 289 rather than 271, flow time decreased from 5 to 13 days and there is profit now, ROI %16. But is it worth it? Perhaps it would be more accurate to reduce variability by improving the process rather than stock. Let's try.
Let's tackle variability: We're back to the initial scenario, 100 days, one-piece flow, working 6-sigma. We were able to reduce the variability in the last (5th) process to the 3-4 instead of 1-6. Will it be enough? While we were expecting 350, we were able to produce 171 units (155 units in the original), the flow time decreased from 30 days to 28 days, and WIP decreased from 105 units to 95 units. We're at a loss, ROI -42%. It was not enough to improve a single process.
Let's fight variability more: Let's improve the first process and one more middle process (3 of the 5 processes in total), let's continue with the flow of one-piece, we are doing Lean-6-Sigma. Production increased from 171 units to 223 units, flow times decreased from 28 days to 21 days, and WIP decreased from 95 units to 71 units. We're at a loss, ROI -32%. So the one-piece flow is not for us...
Let's challenge the variability with WIP: We try by adding as many WIP as the number of stations, namely 5 units. Production increased from 223 units to 323 units! Flow time decreased from 21 days to 12 days, and WIP decreased from 71 units to 40 units. We made a profit, ROI 145%! We are very tired, but I guess it will be worth it. Let me talk to the supervisors in production...
"Let's just fill the line" - Supervisor's opinion: We are back to the original, we will work for 100 days. The variability of the processes is in the range of 1-6 as usual. I agreed to delay the orders for 1-2 days, and supervisors filled the line to reach 10 pieces of WIP between the processes. Production increased from 155 units to 318 units, flow time decreased from 30 days to 10 days, and WIP decreased from 105 units to 60 units. We made a profit, ROI 90%. We've had a pretty brilliant result, just agreeing to delay without dealing with consultants, improvements... Isn't it weird?
"Let's improve all operations" - Accountant's opinion: We are back to the original, we will work for 100 days. Improved variability of all processes to range between 3-4 instead of 1-6. We will continue with one-piece flow. Production was 302 units, flow time was 10 days, and WIP was 30 units. We made a profit, ROI 49%. Indeed, too much effort for too little gain, it was not worth it. We've already done better than that, with less effort, as the master said.
Let's read the GOAL book: Let the output (E) process be our bottleneck, we want to have a continuous production with 3.5 units. We will arrange a WIP of 18 pieces in total, with 3.5 pieces each day and 5 days for 5 processes so that it will never starve. We will locate 10 units of WIP before E, and distribute the balance between the other processes. A, B, C, and D are non-bottleneck processes and they have protective capacity. Production increased from 155 units to 348 units, which is almost our target, the flow time decreased to 5 days as we had targeted, and WIP was almost exhausted as we had targeted and decreased to 15 units. We made a profit, ROI 281%!! We didn't spoil the production area into a warehouse, no need for consultants and our flow time made us the fastest supplier on the market! We achieved 4 times the profit and 12 times the ROI of our plan.
Let's combine the GOAL book + simulator: The rule of thumb for protective capacity is simply 30% more than bottleneck capacity. Since the bottleneck is 3.5 units, we fix the excess capacities with 4.5 units. The production line now looked like a balanced line, but is still unbalanced concerning demand, with more-than-demand capacity. Excess capacity is not a waste, it is protective capacity. Let's improve the process at the bottleneck, bringing its variability from 1-6 to 3-4. Let the WIP be as much as 5 days of production, 18 units, 10 of which will remain before the bottleneck. Let's try: 347 units produced, 4 days flow time, WIP is 10. We made a profit, we reached ROI 51%!
Let's replace the bottleneck with the GOAL book + simulator: When the bottleneck is C instead of E, we can now take 3 days instead of 5 days therefore the WIP will be 11-12 units. Let's try. 348 pieces of production, 3 days flow time, and 7 pieces of WIP. We made a profit, ROI 772%!!! When the bottleneck is at the end, on-day delivery improves, but flow time and WIP increase. A more balanced view is obtained when the bottleneck is in the middle.
Let's transfer the bottleneck from production to sales with the GOAL book + simulator; S-DBR: When the bottleneck is A instead of E, it means that there is no problem in production anymore. You may think that we are not getting enough orders. To overcome the problems of one-piece flow, we distribute 6 random WIP equally. Let's try. 347 pieces of production, 4 days flow time, 9 pieces of WIP. We made a profit, ROI 610%!
Let's get the bottleneck (E) over with the GOAL book + simulator, cancel the premium: Customers are not willing to pay more when we deliver on the day but they like to penalize when we are late. So let's try it by canceling the target-achieving bonus. 347 pieces of production, 3 days of flow time, 9 pieces of WIP. We made a profit, ROI 260%. We earn money even if there is no bonus, and we will no longer be penalized.
What if the demand is not stable?: So far, we have received a constant 3.5 units each day. Now let's try by considering 3.5 units with a uniform distribution (in the range of 2-5). 343 pieces of production, 4 days of flow time, 10 pieces of WIP. We made a profit, ROI was 499%. The system works even when demand fluctuates. While producing at this speed, we can now consider MTO (manufacturing to order) products for which we produce MTS (production to stock).
Let's look at the scenarios: There are 5 scenarios. We can compare them by installing and running them at the same time.
1. Chaotic-Balanced-No intermediate stock scenario: The capacities and variability of the 5 stations are the same, there is no stock between them (one piece flow), demand = plan = 350 units
This simulator is like a teacher, I loved experimenting and learning with it.
I saved 3 scenarios of my own choosing and compared them with the pre-loaded scenario #3.
Reduced variation-buffers=5: In this scenario, all processes have 3-4 variances and there are 5 units of stock between them. (Engineer's approach)
Initial variation-buffers=20: In this scenario, all processes are variable 1-6 and there are 20 units of stock between them.(Supervisor's approach)
No variation-buffers=0: In this scenario, the last bottleneck is 3-4 and the others are 4-4 variances and there is no intermediate stock. (Accountant's approach)
CONCLUSION
Line planning based on traditional capacity balancing does not yield the expected result in a typical production line that is interconnected and variable. It does not improve over time and stockpiling in between or trying to reduce variability in all processes at the same time hurts cost and flow time.
Theory of Constraints DBR solution seems to be the best solution with its simple structure, practicality and good results in a short time.
