Smart CIP project in Germany
The goal of a project of the Industrial Community Research (IGF) in Germany is to develop a self-learning automation system for resource-efficient cleaning processes. The IGF project has been carried out by scientists from the Fraunhofer Institute for Process Engineering and Packaging and the University of Erlangen-Nuremberg.
An economical cleaning method
Cleaning a production system can be done cheaper and more environmental friendly with artificial intelligence, according to the Cleaning 4.0 project from the German universities.
In order to avoid contamination of food the cleaning of kettles, tanks, dosing systems and other processing machines undergo a cleaning-in-place (CIP), where a cleaning process without disassembly of the equipment is carried out, which is labour intensive and takes a long time. Water, cleaning agents, energy and time are used in excess. The cleaning-related downtime in the food industry averages 15-20% of total production time.
The cleaning process thus has considerable economic relevance, the costs of which often exceed 10% of the total production cost of a food. Approaches to control the CIP processes more precisely and more appropriately have not existed so far.
Making CIP an intelligent process
Due to the complex nature of the problem, the system was initially developed to demonstrate its technical feasibility. A 10,000-litre stainless steel tank was used, for which CIP cleaning uses motor-driven jet stream cleaners. The pump capacity or the volume flow of the cleaning water was selected as a parameter for the self-learning control; all other parameters such as the temperature or the speed of movement of the jet cleaner were kept constant. As a pattern, various food products such as ketchup, mayonnaise, mustard and two pudding products were used.
To measure the inline cleaning, the scientists used optical pollution sensors that visualised residual contamination via the fluorescence method, thus enabling real-time monitoring of cleanliness. The sensors were applied in the tank in such a way that particularly hygienic critical points such as agitators or level rings were in the field of vision.
In the measurement, the emitted UV light excited certain components such as proteins or oils for fluorescence, whereby they emitted light in the visible range. This light was then captured by a camera built into the sensor and analysed. For pattern recognition in the camera, recordings and for the evaluation of the data artificial neural networks (KNN) were used. The measured data recorded the cleaning progress and reported this to the jet cleaner.
In the test series, the jet cleaner then adjusted its pump performance as needed and learned over several cleaning cycles, so that the cleaning was optimised step by step.
During the step-by-step self-adjustment of the cleaning parameters from cleaning to cleaning, the system linked occurring events (such as a bottling stop) and evaluated the results of parameter changes made. Adjustments not improving the cleaning process were discarded and successful parameter changes accepted.
Through a simulation, this learning process should be shortened. However, until all relevant process variables can be taken into account in a complex cleaning simulation, further research is required.
The results achieved in the project for the optimised use of cleaning resources already show great potential. Compared to a standard cleaning, the intelligent process control enabled 35% of the cleaning water (in case of lost cleaning, ie if all media are disposed of after the respective cleaning step) and 55% of the pump energy was saved.