|
| Algorithm used | AI model taxonomy | Depth layer sizes, training time, and testing time | Dataset | Framework, core language, and interface | Advantages | Disadvantages | Ref. |
|
| Xception and Inception V3, d Resnet-50 | Unsupervised | 71 layers deep. Twenty % testing and eighty % training data. | ImageNet dataset | TensorFlow
Python language | Imaging models may be helpful to come across faults in injection molding, along with weaving, which is tough to become aware of with the naked eyes. Model testing performs the best at detecting short forming, with good precision, recall, and high F1 score. | It is required to make a dedicated neural network to bring images and the sensor data that could carry required information on the internal faults. | [73] |
| SMOTE algorithm is chosen as the training algorithm | Supervised | The cycle time is 21.84 seconds, and inject time is 1.35 seconds | The dataset is imbalanced and stored in a NoSQL database, and preprocessing is done using significant data methods | An extensive data framework is used, and machine classification and prediction models are used | The faults are estimated before occurrence and also provide an expert interpretation that can give insight into the main problem of the mistakes | Quality rates are very excessive for most types of synthetic parts.
Much data are required before the overall performance of the model can become handy. | [74] |
| Artificial neural networks (ANNs) and decision tree | Unsupervised | Has three layers (hidden layer, input layer, and output layer), where twenty two neurons were on a hidden layer | Data include forty-one machines and process parameters from one hundred and sixty machine runs based on difference about holding of pressure time, holding off pressure, back pressure, injection speed, cooling time, barrel temperature, screw speed, and temperature of the tool parameter | Python Language and classification model | The model can differentiate good or bad parts with 99.375% accuracy | No chance to categorize the type of defect that occurs | [44] |
| Cluster analysis | Unsupervised learning | NA | Datasets related to the molds are used in the production | NA | Finds the molds wearing out and predicts in real-time which has an excellent early warning system | Complete data are not fed into models | [72] |
| Control algorithm | Unsupervised algorithm | NA | PVT (pressure, specific volume, and temperature) data are used | Plotter for plotting the curve and graph | Consistent quality of the molded part is maintained | Many training data are required to be acquired | [75] |
| AI models are used | Supervised learning | Cycle time is 37 seconds, and inject time is 3 seconds | Data are generated from many injection molding cycles by using temp sensors and pressure caused by the cavity. | Python language | Implements the control system for good quality for product generated by the injection molding through real-time monitoring. | The whole artificial intelligence-based control system is not ready | [76] |
|
