The partial least squares (PLS) algorithm is generally suitable for the fault detection of industrial processes under stable conditions. In increasingly complex industrial processes, process data does not satisfy the normal distribution and there are nonlinear, dynamic, and multimodal problems. For the multimodal problem, a large number of modality discrimination methods are available. But these methods do not consider quality-related factors, and therefore these methods do not apply to quality-related algorithms. For this reason, this paper proposes a new quality-related modal differentiation rule for quality-related algorithms. This rule uses kernel fuzzy clustering to perform preliminary clustering on the increase data for linearly increasing time variables, and then further quantifies the modality accurately by the quality related indicators. At the same time, the complexity of the process makes the static control limit unable to meet the requirements of fault detection. The applicable range of the existing dynamic control limit has certain limitations. By improving the dynamic control limit, this paper extends it to the generalized dynamic integrated control limit. In the experiment, firstly based on two nonlinear partial least squares models, the proposed methods are applied to the fault detection of penicillin fermentation process, which greatly reduce the false negative rate and false positive rate. Finally, a numerical simulation experiment verifies the rationality of adding a linear incremental time variable.