Abstract: Under the deep empowerment of artificial intelligence and large model technologies, data-driven approaches have become the core model for industry decision-making. High-quality datasets, as critical production factors, are essential for systematic management. Addressing the current lack of comprehensive lifecycle planning in industry dataset construction—which leads to challenges in ensuring dataset quality—a methodology inspired by mature software lifecycle management is applied. Industry datasets are treated as specialized products, subject to full-process control across three stages: planning, development, and maintenance. This encompasses seven phases including feasibility analysis, requirements analysis, architecture design, dataset development, quality evaluation, model validation, and operation and maintenance, with defined objectives, core tasks, and standardized deliverables for each stage. Additionally, considering the dynamic evolution characteristics of datasets and the ever-changing industry demands, an agile dataset development life cycle (DDLC) model is established. Through short-cycle iterations, cross-role collaboration, and closed-loop optimization, this model enables continuous delivery and dynamic refinement of datasets. Validation is conducted using a case study on the construction of a high-quality process dataset for a rail transit equipment manufacturing enterprise. The results demonstrate that the model effectively standardizes the entire dataset construction process, significantly enhances the performance of process-oriented large models in tasks such as generating process instructions and compliance verification, and provides a scientific and practical framework for the construction and management of high-quality industry datasets.