There have always been two ways to implement APS: one is a standalone APS system, which incorporates algorithms from operations research, intelligent optimization, and even big data and artificial intelligence. While not a traditional analytical solution, it provides a relatively accurate scheduling method. The second method uses simulation to schedule production, aiming to replicate the internal adaptive and self-consistent mechanism to obtain the scheduling result.
Initially, these two forms developed largely in parallel, but now, facing new production scenario demands, they are actually moving towards a near-complete break or a kind of fusion. The basis for this break or fusion can be seen in the 2019 US technology embargo against China, one aspect of which was logistics system modeling and simulation. The US embargo was certainly not aimed at the 3D display technologies commonly used in logistics modeling and simulation, but rather at the simulation glasses behind them; and the essence of this simulation engine is actually an APS (Automatic Product System).
My research on APS actually began with simulation. The Digital Manufacturing Institute at Beijing Institute of Technology, where I work, started researching production line modeling and simulation in 2003. We were probably the second university in China to introduce TECHNOMATIX eMPlant (the predecessor of Siemens' current PlantSimulation). At that time, our core work was secondary development, which involved developing various operating rules or production scheduling rules. However, we quickly abandoned the simulation platform and developed our first independent APS system around 2008.
There are three reasons for abandoning the simulation platform and forming an independent system.
Firstly, there's a desire to avoid being used on other people's platforms, especially overseas ones, and there's also an element of strengthening self-reliance and control.
Secondly, simulation has traditionally been used because the entire production or manufacturing system is extremely complex, making it unsuitable to establish analytical solutions or stochastic processes similar to Markov models. Therefore, it is hoped that the results can be obtained through simulation. However, with the deepening of APS research, the logic of simulation can be abstracted and refined to establish an algorithmic process. Therefore, it is unnecessary to do this based on a simulation platform. Conversely, manufacturing and production systems are not yet so complex that they necessarily require simulation for operation and solution.
Thirdly, and most fundamentally, simulation engines have significant limitations, especially since they are primarily based on heuristic rules. This makes them severely inadequate for optimizing production scheduling results; otherwise, a series of intelligent optimization algorithms wouldn't have been developed. This is an inherent constraint of simulation, built upon a simulated operational mechanism, and it's very difficult to overcome. Furthermore, the customized solution constraints of different industries or enterprises require the APS's inherent algorithmic model to have good extensibility, while simulation platforms actually introduce many limitations that hinder development.
Fourth, the visualization of simulation operation effects has now been largely achieved through digital twins of workshops or production lines, further reducing the need for traditional simulation applications.
However, with the further development of APS, I feel that it once again confirms that the development of technology and things is a spiral ascent or a negation of negation, which is very much in line with the philosophical laws of development.
First, the internal operating mechanism of simulation inherently possesses intelligence. If you are familiar with the 4th generation APS, it is basically a simulation operation mode, but it is not the traditional simulation approach; it requires improvement, but the essence is the same.
Secondly, in some special application scenarios, such as AGV path planning and refined control execution, the simulation mechanism is still very good, even though it does not use a traditional simulation platform on the surface.
While APS and simulation may appear to be diverging on the surface, they are actually converging at a fundamental level; the external form is no longer important. This convergence represents a convergence of the abstract logic of manufacturing system operation. However, in terms of flexibility, APS is easier to control than simulation and is transforming traditional simulation systems. I have already begun organizing and planning the development of a software system that integrates APS and simulation, and a prototype should be available this year.
