In complex manufacturing systems, particularly in continuous and cascaded process manufacturing systems with cascading processes, decision-making operations such as parameter adjustments are usually built on a pre-defined rule base. Such a decision-making paradigm results in poor scalability and adaptability of manufacturing systems in dynamically changing environments. However, these expert systems are usually built on a limited and fixed rule base, resulting in poor scalability and adaptability, making it difficult to meet the demand for efficient decision-making in dynamically changing environments that require extensive knowledge. Large language models (LLMs), by efficiently compressing vast amounts of information, can perform high-quality dialogue and response tasks in general domains, showing the potential to become a new type of expert system in industrial settings. Effective application of LLMs relies on a large number of prompt word engineering, which usually requires manual construction. This paper proposes a method for constructing Chain of Thought (CoT) prompts guided by industrial knowledge. By abstractly describing the decision-making process model, this method forms multi-step reasoning guidance information, thereby improving the response quality of LLMs in assisting decision-making within cascaded process manufacturing systems. The proposed method demonstrates the feasibility of using general LLMs for decision-making support tasks in the manufacturing domain.