Product time estimation for warp machine in weaving enterprises

Since the processing time has a significant impact on the key performance indicators of the planning process such as total tardiness, total flow time, and completion time, its correct determination has a key role in the effective management of the process. Traditionally, processing time is determined by time study or simple calculations and is assumed to be precisely known in advance of planning and is usually a fixed value. In some cases, the function used to calculate it is known in advance, although the processing time itself is not known in advance. Such situations often occur in chemical and metallurgical processes. Due to some unexpected situations in the actual production process, the processing times cannot be recorded exactly. It is seen that this situation, which is called uncertainty in processing times, is handled in different ways in the literature. The first of these is the fuzzification of the processing time. A second approach to modeling indefinite processing time is interval numbers. Random or stochastic processing time is another way of expressing uncertainty in processing time. In this approach, the processing time of each job is a random variable with a known probability distribution. In this study, the uncertainty in the processing time is not caused by probabilistic changing factors such as machine failure, power outage, but by a combination of factors with previously known fixed values. In this study, inspired by the warp preparation operation, which is one of the important components of textile production processes, supervised machine learning approaches are used to estimate the processing time of the warp preparation operation. The processing time involved in the preparation of beams for the warp yarns, which are used in weaving machines, is subject to variations based on multiple parameters. To address this, interviews were conducted with subject matter experts, resulting in the identification of 11 features that could potentially impact the processing time. Subsequently, a total of 12 machine learning algorithms, comprising 6 linear and 6 nonlinear models, were employed to estimate the warp preparation process time. The experimental study utilized real-life data extracted from the enterprise's ERP system, ensuring its practical relevance.12 different supervised machine learning algorithms were run with both training and test datasets and the results were presented in a comparative way. It has been observed that boosting algorithms are superior in terms of both training/adjustment time and estimation accuracy.