The knowlEdge project carries out 3 pilot demonstrators (and 4 use cases) in order to verify the knowledge architecture. The four use cases have been selected from distinct manufacturing and process industries: food, plastic-car parts and gear-car-machinery.
For what concerns the food industry the pilot demonstrator is Parmalat S.P.A. part of the Lactalis Group, specialized in milk, dairy product and fruit juice production. In particular, one of its use case regards the scheduling of the production for the Collecchio site.
Collecchio was Parmalat’s first production facility and currently it is the Headquarter and the plant that still produces the largest volumes of goods. Nowadays the plant produces more than 380,000 tons of finished product per year that includes UHT sterilized milk in TetraPak and bottles, juices, yogurt and milk derivatives and employs more than 400 people.
The objective of this use case is to create a scheduling tool that is able to:
– Optimize production sequences respecting constraint and limits of the entire production process
– Improving KPIs, such as OEE (overall equipment effectiveness)
– Minimize costs
Since the beginning of the project, many steps have been made towards this goal.
In the first place, it was outlined and identified the target through a survey. The scheduling of the yogurt department was selected as the core of the pilot.
After the target identification, a key part of the pilot was to enable the partners to understand the yogurt process and the structure of the Parmalat database. In order to allow the partners to experience almost first-hand its production reality, Parmalat modelled the milk process in the Collecchio production plant from the arrival of the milk to the final product. With the help of the yogurt production planner and the department manager, the possible optimization and the constraints were identified and defined. The result of the modelling was a virtual tour regarding general milk matters as well as details about specific productions shared among the consortium during a workshop on the 14th of March 2022.
In May 2022 the first version of the scheduling simulator was released. The Discrete Event Simulator (DES) is a model of a system processing system events as a sequence in time. In the project, the simulator is used in context of yogurt production demonstrator. The inputs of the simulator are the constraint tables that indicate on which resource(s) each product can be processed and what is the processing capacity on the resources for each product.
The main objective in creating a schedule is to generate clusters of the products where the product properties are as similar as possible; also, it is important to order the products within the clusters minimizing the setup required. E.g., change from fat free strawberry yogurt to lactose free yogurt containing no fruits requires washing off the fruit and lactose residues and change of the raw yogurt feeding tank.
The clustering in the simulator is done by rulesets, which are utilized in search for the best match on each product change occasion on each resource. A ruleset is a sequence of selection rules where the first rule matches most of the properties, e.g., same raw yogurt, fruit and package, the next rule then looks for same raw yogurt and fruit, the rules are then continued towards the rule with least matching properties. The rules are tested against the demand from best match towards end until a matching product is found. Utilizing the rules this way automatically generates clusters of similar products from the demand data. Any number of new clustering rule sets can be defined, the simulation can be then run with alternate rule sets to generate different schedules for comparison of the schedules and their KPIs with other schedules. See the video below for an illustration of the functioning of the Simulator.
The simulator is currently being tested and updated. Within the knowlEdge-platform the simulator is used as an external component for generating numerous alternate production schedules to use as machine learning data sets for the verifiable and integrative AI. So, the next step will regard the optimizations of production sequences with the help of AI technologies.