KI Use Cases in der Produktion: Losgrößenoptimierung
How can AI help find the optimal batch sizes for cost-effective multi-variant production? We explore this question in Part 3 of our mini-series on AI Use Cases in Production. Each of the presented AI use cases builds on the 4 Opportunities for Value Creation with AI in Industry.
Determining the correct batch sizes for daily multi-variant production is an art, even when customer requirements are fundamentally known. This is especially true when the offered product variety is high and customers occasionally need smaller quantities. Once machines or equipment must be reconfigured for production of two different product variants, the number and size of batches to be produced determine their availability and the required setup effort. Batches can be enlarged by combining identical variants from different customer or product requirements, or by producing for inventory. Large batches are generally good for productivity, but can lead to excess inventory of certain variants or insufficient machine capacity for smaller or urgent demand quantities. High capital tied up in inventory or customer delivery issues can be the consequence.
AI can help determine appropriate batch sizes to best meet customer requirements, delivery deadlines, and inventory levels. This AI use case will be built up step by step in the following.
An MES Paves the Way for AI
The starting point for AI development is an AI-suitable data foundation. The necessary information for batch size optimization is diverse and scattered throughout the factory. This includes, among other things, the product or variant master data, inventory levels, customer requirements with article, quantity and delivery date specifications, currently available machine and personnel resources, as well as available tools and production materials. All this information is needed along with corresponding operational data for a period as far back as possible.
With an appropriate data strategy, alternative data stores for the required information can certainly be implemented, but in practice, manufacturing companies typically rely on or already have a suitable Manufacturing Execution System (MES).
Stage 1: AI Recognition
A recognition AI can search the data for common batch size compositions and differentiate them from each other. The resulting groups can then be categorized by process experts, for example "productivity maximization," "small-batch production," or "50:50." The result can already be used in an advisory system that displays the category of the current batch size composition or its deviation from previously known categories.
Stage 2: AI Diagnosis
In the next step, a diagnostic AI links batch size categories with additional operational data, such as inventory levels, delivery deadlines, overall capacity utilization, or personnel availability. Here, the AI should particularly learn which constellations of batch sizes and operational conditions have led to problems with inventory levels, personnel requirements, and delivery reliability. Occurring problem cases can thus be systematically compared against known problem patterns and can continuously uncover improvement potential.
The way AI learns, stores, and outputs relationships is often incomprehensible to humans, since simple "if X, then Y" relationships rarely occur. AI diagnoses can therefore be unsatisfying for humans without appropriate post-processing. aiXbrain therefore works intensively on human-centered AI that provides human-understandable explanations, among other things.
Stage 3: AI Prediction
A predictive AI reverses the relationships learned in AI diagnosis and draws conclusions from selected batch sizes and observed operational conditions about the likely impact on critical operational parameters such as delivery deadlines or personnel requirements. Through a proactive warning system, production planners can receive information from the predictive AI already in the planning phase about what problems are likely to arise based on the selected planning parameters.
Stage 4: AI Planning
In the final expansion stage, a planning AI, in case of a warning from the predictive AI, presents production planners with a batch size suggestion that best counteracts the emerging problems. The planning AI draws both on batch size-operational state pairs observed over time and on countermeasures implemented by planners. This knowledge of the planning AI, which becomes more comprehensive over time, can potentially contribute to significant savings in costs, time, and effort.
Value of AI for Batch Size Optimization
The analysis of the effects of certain batch size compositions and their optimization exceeds human capacity beyond a certain product offering. Too many intertwined side effects make systematic or even manual processing impossible. AI in its various forms of application brings light to areas where humans, even with the help of conventional software, cannot see, and ensures sustainable improvements in operational processes and customer satisfaction.
The collaborative interplay between production planners and AI across the various expansion stages also conserves valuable know-how in digital form. This knowledge is sustainably available to all current and future employees.
If you are interested in further value-creating AI use cases in production, we recommend the other articles of our mini-series:
