Cognitive Production Planning
Dynamic production planning
KAYROS is a technology for dynamic production planning. It plans any order pool on the most complex production networks with the aim of minimizing setup costs, meeting deadlines, using existing tools efficiently and much more. The underlying AI technology enables KAYROS to achieve very high calculation quality within very short calculation times.
The challenge of production planning
In industrial production, planning is often still done manually – for example, when which job is to be produced on which machine and how. This approach is increasingly coming up against its limits in today’s world, as production must become more flexible and faster. The result is that delivery dates are not met, set-up costs are unnecessarily high and optimization potential is unused.
So far, automatic, algorithmic approaches could only provide a limited remedy, because they are either too slow or too inflexible.
The solution is called KAYROS
The KAYROS technology developed by PerfectPattern enables dynamic production planning – at lightning speed and with maximum precision.
KAYROS plans any number of job pools, even on complex production networks, with the aim of minimizing setup costs, ensuring that deadlines are met and much more.
The special capabilities of KAYROS lie in its ability to solve non-linear decision problems with a very large number of conflicting parameters in a very short time.
In the optimized production planning, raw material costs are weighed against setup costs. The result is a compromise that minimizes both setup costs and raw material consumption.
KAYROS schedules the complete order pool in a very short time. As a result, potential capacity bottlenecks become visible very early on and can thus be avoided. This creates scope for decision-making and ensures that deadlines are met.
Thanks to KAYROS’ short computing times, even for very complex production scenarios, it is possible to react agilely and comprehensively even to short-term changes in production. Thus, even when things get a bit tight, good decisions are still made.
KAYROS has its own language for modeling and describing virtually any machine. This flexibility allows KAYROS to model even the most specific machine engineering requirements.
KAYROS controls the efficient use of special production tools, which are only available in limited quantities and are needed at different times at different points in the production network.
KAYROS knows the exact structure of the production network. This enables it to take transport routes into account and ensures that the transport distances of materials between the machines are also minimized.
KAYROS itself finds the most efficient production for its products. It independently considers all possible production alternatives that meet the requirements of the product and picks the most efficient.
The KAYROS Principle
KAYROS models the flow of orders through the production network and calculates the current optimal production plan. Through the underlying modeling, KAYROS knows which products can be produced on which machines with maximum time and cost efficiency. Thanks to its dynamic nature, the system adapts spontaneously to unforeseeable influences such as machine failures and compensates for them by rescheduling.
Modeling of machines
In KAYROS, each machine can be modeled in such a way that all parameters relevant to production planning are taken into account. This is based on the following aspects:
- A machine can be put into different states by changeover procedures. This can also mean switching certain modules on or off or removing them. A changeover procedure costs time, money and possibly raw material resources.
- In any state the machine can perform certain processes. Each process costs time and money. For the calculation, the processes can be linked to aspects such as the number of pieces per hour, an area per time or other measurable variables.
- Every process that the machine performs fulfills certain requirements that a product has.
- It is also possible to specify the transport routes between certain machines or machine parts that KAYROS should consider.
By modeling a machine, KAYROS creates a digital image of the real machine, which has the same capabilities as the original and generates the same costs. Now KAYROS needs the information when the machines are available.
Equally important is the continuous feedback of each machine to KAYROS about its current condition. This enables KAYROS to react accordingly if, for example, a machine breaks down and switch to other, possibly more expensive capacities. Here the rule is: the more up-to-date the information, the better the result.
Modeling of products
Each product is recorded in KAYROS as the sum of individual production steps. In each production step, the product or part of the product is brought into a new state by a process that is closer to the final product. Each product or product component has specific requirements which have to be fulfilled during the production process and often in a specific order.
Another input for KAYROS are the orders. As soon as a new order is received, it must be transmitted to KAYROS. KAYROS will then schedule this order according to its priority and adjust the production schedule accordingly.
The decision making process
KAYROS has a powerful decision making technology. Based on global target functions, it makes the relevant decisions for dynamic production planning quickly and precisely.
KAYROS follows the concept of “Prescriptive Analytics”. In contrast to “Predictive Analytics”, this means that not only predictions about expected behavior are made, but that decisions are made that move a system in a desired direction.
The special capabilities of KAYROS lie in its ability to solve decision problems in a very short time, which are non-linear and contain a very large number of conflicting influencing parameters.
The learning system
In the industrial environment, continuous methods can usually not be applied – the decision space is incredibly large, and in addition, the objective functions show massive leaps even in small areas of the decision space. KAYROS explicitly constructs the next decision from the learned objectives in order to achieve the given global goals. This process is called Constructive Reinforcement Learning
KAYROS uses input data and functional relationships that define the system to be calculated (the factory). These are for example orders, priorities, machine capabilities and speed etc. or also probabilities of occurrence of certain events like machine failures or similar. Based on these data and models, KAYROS uses Constructive Reinforcement Learning to calculate a production plan. This plan contains precise instructions for controlling a factory.