In the discrete manufacturing world objects are usually produced in a finite sequence of steps on dedicated work stations organized into an assembly line. A batch of objects waiting for production or assembly must therefore be assigned first to a production line and second to a position within the sequence on the line.
The set of KPI’s is usually specific to the process in question. However, the following quantities are usually to be found in most industries:
- Production slack, i.e. the difference between cycle time and productive work content per cycle
- Setup time, i.e. the total of non productive work arising from changing types of objects per work station
- Timeliness, i.e. the total amount of time where effective completion time exceeds scheduled completion time
- Work load peaks, i.e. the amount of variance per work station from one cycle to the next due high frequency in model changes
Target function summarizing the overall penalties in a sequencing problem
Minimizing slack times for an assembly line, snapshot across work stations
As can be seen from figure 3, the overall penalty function can be drastically reduced by using Local Search Optimization Heuristics. In this case it is “Simulated Annealing”, which does the trick.
In case there is more than one assembly line, the result of a successful sequencing run is
- A simultaneous partition of subsets assigned to each line and
- A permutation for each subset
In case there is just one assembly line, the result of a successful sequencing run is simply a permutation of the number of objects. Figure 4 below shows, how overall slack times on a production line can be addressed and minimized through optimal sequencing. Figure 4 is a snapshot for a given cycle across the stations of a line: