The technological promise is there: with computing power, and why not artificial intelligence, your production workshop, however complex it may be, will be scheduled in real time.
The intelligent algorithm knows how to plan tasks, optimize setup times, ensure sequences, and take constraints into account—you will be able to promise optimized delivery dates and keep those promises, with a workshop managed with precision in real time.
It sounds appealing, but in the vast majority of cases, it doesn't work.
If your production process is fairly simple—assembly lines, for example—it may work reasonably well. But in this case, one might question the value of sophisticated algorithms—a repetitive sequence and a simple planning wheel are probably sufficient to organize the work.
If you are in an environment with small batches of diverse products and multiple operations, it will be more complicated... and chances are your teams will try to figure out how to prioritize using Excel, possibly several times a day.
What are the causes of these difficulties?
Uncertainties are, of course, part of the equation. Machine breakdowns, absenteeism, quality issues, shortages.
But that's not the only problem. There is, in fact, a structural problem. All algorithms work based on data. And what data is needed to carry out scheduling? Routings. A sequence of operations and associated times.
We're between us, so you can admit it. How good are your routings?
These sequences define setup times, operating times, possibly technical times, and subcontracting lead times.
If you expect your scheduling tool to tell you when to do what on which machine, knowing that you may have several thousand planned routing operations in your factory, you need very high-quality data—and therefore you need to trust your routings.
Let's take the example below. Your magic algorithm has determined that operation 700 will start on January 8 at 8 p.m. and end on January 9 at 8 a.m. What are the chances of this happening? Probably zero—because the actual operation times and setup times for this work order and other orders that go through the same resources will not exactly match the planned range times.
These range times are estimates, so your "optimized" manufacturing schedule is also an estimate.
Never mind, you might say. Of course, our schedule will not be met. There will be delays here, perhaps advances there. So, all we need to do, thanks to the available computing power, is to reschedule the remaining steps at any given moment.
Rescheduling means reshuffling the deck, changing sequences and priorities. But to be efficient, a factory needs stability. And to be respected, it would be good if the dates promised to your customers weren't constantly being changed by an algorithm, right?
Since our scheduling is roughly correct but not precise, we will insert time buffers—safety nets—at key stages. In our example, we insert a 24-hour buffer at operation 700 and a 16-hour buffer at the end of the work order.
If the times are not respected from operation 100 to 700, no problem: operation 800 is protected by a 24-hour buffer (a queue). There is no need for permanent rescheduling, you just must respect the priorities on the USI3 resource.
And all operations on this critical USI3 resource are protected:
Not only is the daily life of a factory full of uncertainties, but the basic technical data - the routings - cannot be exact.
For flawless execution, you must therefore accept that you are working based on estimates. Your program must include safeguards—and provide visibility to your production teams so that there is no ambiguity about priorities.
By anticipating the fact that there will be delays, you will confidently be able to deliver on time!