Abstract:In the process of establishing a quality prediction model for injection molded products, challenges often arise due to the inability to fully utilize process data in injection molding, resulting in low model prediction accuracy. With emerging injection molding processes and limited training data, there is an immediate need to address how to use limited data samples for online monitoring and assessment of the quality of different injection molded products and to leverage existing injection molding data effectively. This paper proposed a quality prediction approach for small-sample injection molded products, relying on a double-layer bidirectional gated recurrent unit network (DL-BiGRU), multi-feature fusion and deep transfer learning. Initially, A new injection molded product quality prediction method is proposed, based on DL-BiGRU and multi-feature fusion. This serves as a base model in the source domain for transfer learning, used to extract generic knowledge of the injection molding process. Subsequently, by introducing model-based deep transfer learning, a target domain transfer model is constructed to reuse generic injection molding knowledge. In order to address the issue of non-uniform dimensions of input data in the source domain and target domain, this paper introduced a dimension unification layer into the target domain transfer model. This layer facilitated the training of the transfer model following knowledge transfer. This approach successfully realizes knowledge transfer and reuse of injection molding generic knowledge from large-sample datasets to small-sample datasets. When applied to injection molding datasets of different quality, the transfer model enables knowledge transfer from injection molded product dimension to injection molded product weight. The results show that, in comparison to the conventional base method, the proposed transfer learning method has a higher model quality and provides more accurate predictions, particularly for small-sample datasets.

Contrastive Decoder Generator for Few-Shot Learning in Product Quality Prediction.

An Adversarial Bidirectional Serial–parallel LSTM-based QTD Framework for Product Quality Prediction

Path Enhanced Bidirectional Graph Attention Network for Quality Prediction in Multistage Manufacturing Process

Cooperative Deep Dynamic Feature Extraction and Variable Time-Delay Estimation for Industrial Quality Prediction

Deep Learning of Partially Labeled Data for Quality Prediction Based on Stacked Target-Related Laplacian Autoencoder.

Deep Supervised Dictionary Learning for No-Reference Image Quality Assessment

Deep Learning of Complex Batch Process Data and Its Application on Quality Prediction

A novel DL-BiGRU multi-feature fusion and deep transfer learning based modeling approach for quality prediction of injection molded products using small-sample datasets

Multiscale Cost-Sensitive Learning-Based Assembly Quality Prediction Approach under Imbalanced Data

Coarse-to-Fine Few-Shot Defect Recognition with Dynamic Weighting and Joint Metric

Quality Prediction of Injection Molding Process Using CNN and Parallel-Cascade LSTMs

Two-stage Imbalanced Learning-Based Quality Prediction Method for Wheel Hub Assembly

A Multiphase Dual Attention-Based LSTM Neural Network for Industrial Product Quality Prediction

Quality Prediction of Discrete Manufacturing Process Based on CGAN&Catboost Hybrid Model

Deep Learning for Just-In-Time Defect Prediction

Deep Learning for Industrial KPI Prediction: when Ensemble Learning Meets Semi-Supervised Data

Production quality prediction of cross-specification products using dynamic deep transfer learning network

Few-shot defect detection using feature enhancement and image generation for manufacturing quality inspection

CLG: Contrastive Label Generation with Knowledge for Few-Shot Learning

Quality Prediction on Deep Generative Images

Statistical Prediction of Product Quality in Batch Processes with Limited Batch-Cycle Data