Customers in the manufacturing sector, especially in the automotive industry, have a high demand for individualized products at price levels comparable to traditional mass-production. The contrary objectives of providing a variety of products and operating at minimum costs have introduced a high degree of production planning and control mechanisms based on a stable order sequence for mixed-model assembly lines. A major threat to this development is sequence scrambling, triggered by both operational and product-related root causes. Despite the introduction of Just-in-time and fixed production times, the problem of sequence scrambling remains partially unresolved in the automotive industry. Negative downstream effects range from disruptions in the Just-in-sequence supply chain, to a discontinuation of the production process. A precise prediction of sequence deviations at an early stage allows the introduction of counteractions to stabilize the sequence before disorder emerges. While procedural causes are widely addressed in research, the work at hand requires a different perspective involving a product-related view. Built on unique data from a real-world global automotive manufacturer, a supervised classification model is trained and evaluated. This includes all the necessary steps to design, implement, and assess an AI-artifact, as well as data gathering, preprocessing, algorithm selection, and evaluation. To ensure long-term prediction stability, we include a continuous learning module to counter data drifts. We show that up to 50% of the major deviations can be predicted in advance. However, we do not consider any process-related information, such as machine conditions and shift plans, but solely focus on the exploitation of product features like body type, power train, color, and special equipment.

制造业的客户，尤其是汽车行业的客户，对价格水平与传统批量生产相当的个性化产品有着很高的需求。提供多种产品和以最低成本运营的相反目标为混合模型装配线引入了基于稳定订单序列的高度生产计划和控制机制。这种发展的一个主要威胁是序列打乱，由运营和产品相关的根本原因引发。尽管引入了准时制和固定生产时间，但序列加扰问题在汽车行业中仍然部分未解决。负面的下游影响范围从按顺序供应链的中断到生产过程的中断。早期对序列偏差的精确预测允许在无序出现之前引入反作用以稳定序列。虽然程序原因在研究中得到广泛解决，但手头的工作需要不同的视角，涉及与产品相关的观点。基于来自真实世界的全球汽车制造商的独特数据，训练和评估监督分类模型。这包括设计、实施和评估 AI 工件的所有必要步骤，以及数据收集、预处理、算法选择和评估。为确保长期预测稳定性，我们包含了一个持续学习模块来应对数据漂移。我们表明，可以提前预测多达 50% 的主要偏差。但是，我们不考虑任何与流程相关的信息，