Time-Series Measurement Recording for DTW-Based Quality Prediction

The patent applies preliminary action by performing dynamic time warping (DTW) alignment on historical measurement data before generating the determination model. This preprocessing step aligns the time-series data in advance, ensuring that when the model is used for prediction, the data is already properly synchronized, thus resolving the contradiction between predictive capability and alignment accuracy.

The patent introduces DTW as an intermediary mechanism between raw time-series data and the determination model. This intermediary process transforms misaligned data into aligned data through warping, allowing the model to receive properly synchronized input without requiring real-time alignment during prediction, thereby maintaining both predictive speed and alignment accuracy.

The patent performs DTW alignment as a preliminary action during the offline model generation phase rather than during online prediction. This allows the computationally intensive alignment process to be completed in advance when computational resources are available, reducing the processing complexity during real-time operation while maintaining high alignment accuracy in the generated model.

The patent applies dynamics by using DTW to create a dynamic alignment that adapts to the specific characteristics of each dataset. The warping function is dynamically determined based on the actual time distortions present in the data, allowing the system to handle variable processing requirements while maintaining accuracy, and the aligned model can then be reused multiple times with reduced processing complexity.

The patent implements feedback by incorporating a validity verification step that uses the same determination model to verify its own outputs. The system checks whether determination results fall within expected ranges and provides feedback on model performance, allowing continuous monitoring of model validity while maintaining high determination accuracy through the learned parameters.

The patent performs preliminary validation by generating the determination model with built-in validity checks during the offline training phase. The model is trained to recognize not only accurate patterns but also the boundaries of valid predictions, so when deployed, the model can self-verify its outputs without requiring complex external validation mechanisms, thus maintaining accuracy while simplifying validity detection.