The IPT.Stack lets you regain control over hundreds or even thousands of parameters in constantly changing production environments.
IPT’s secret is its special way of statistical modelling of manufacturing processes in terms of an information theoretical graph. Based on this graph and those manufacturing parameters, that cannot be changed anymore, the IPT.Stack calculates optimal settings for all parameters that are still adjustable. Thereby, it heavily depends on the customer how “optimal” is defined. The optimization goal usually is a balance of particulate interests, like material costs, scrap rate, energy consumption, or cycle times. The IPT.Stack can optimize manufacturing processes as a whole and therefore reduce start-up times significantly. And since parameter recommendations are calculated for each part individually, changes of the environment are taken into account automatically. The manufacturing process becomes much more stable. Especially since the IPT.Stack always tells you how certain its predictions are.
Get an insight of what we can do for you
We Get the Job Done
Introducing data science in manufacturing plants is not an easy task. IPT developed a sound approach consisting of the following steps:
Structuring the Use-Case
First, the use case gets structured, usually within the framework of a workshop. This includes identifying the most promissing optimization goal: e.g. avoiding a certain defect characteristic, reducing the amount of a certain material, energy consumption, or a combination thereof. Subsequently, one identifies those manufacturing processes the chosen optimization goal depends on. All knowledge related to those processes is collected; e.g. known physical interdependencies, or knowledge of experience which combinations of manufacturing parameters yield the best results. Finally, the related data gets assessed. Which data of which quality is available at what cost?
Evaluating the Optimization Potential
Or: Could we optimize production if the world was exactly like we assume it to be?
Based on the previously structured knowledge, a model of the production process is created with the IPT.Stack. It must then be checked whether the compiled use-case offers potential for substantial improvement. First, the model is used as a simulator to generate artificial data. Second, it is tested whether the structure of the model and the amount and quality of the data are sufficient such that the model’s statistical degrees of freedom can be trained with the data. Third, it is examined how good the recommendations for optimal production settings provided by the IPT.Stack are.
Real-World Test with Manually Selected Data
While evaluating the optimization potential, usually a first set of real production data related to the chosen optimization goal can be built. This data is used to test the process hypotheses underlying the statistical model for self-consistency. Furthermore, the data is used to calculate preliminary recommendations for set points for those machines that are related to the optimization goal. The process expert in charge checks the recommended set points and then implements them in production. By this, one formalizes how accurately the statistical model describes production reality .
The final step is to fully integrate the IPT.Stack into the customer’s IT-landscape. The manufacturing parameters as well as the quality data are automatically provided to the IPT.Stack, which in turn feeds back the optimal manufacturing parameters. In addition, process engineers in the shop floor are given access to the graphical user interface of the IPT.Stack. For these system integration tasks, we work closely together with our partner company DE software & control GmbH.