|
Algorithm used | AI model taxonomy | Depth layer sizes, training time, and testing time | Dataset | Framework, core language, and interface | Advantages | Disadvantages | Ref. |
|
Regression and neural network | Supervised learning | Three-layer MLP architectures | Dataset comes from inline industrial measurements | Scikit-learn library, Python, | Shows the best performance about one and only LSTM layer architectures | Recurrent network architecture was also complex | [113] |
The real-time adaptive control algorithm | Supervised learning | Three-layer neural network | NA | Computer simulations | Proposed control strategy is effective under setpoint change and bounded disturbance | For example, a considerable fixed learning rate was considered, so RNNM might suffer from instability | [114] |
Training algorithm | Supervised | Two-layer neural network | Dataset consists of plasticizing stroke, holding pressure, injection velocity, metering stroke, mold opening speed, RPM, injection pressure, cooling time, cushion, and nozzle temperature | Computational tool (DPTNN) | Online monitoring and diagnosis system on false detection of the injection molding machine | Only single network architecture was used | [115] |
TDNN and Elman networks | Supervised learning | NA | NA | MATLAB | TDNN exhibits lesser train time comparatively for the wanted performance | Temporal dependency is large | [116] |
Genetic algorithms | Supervised learning | Three layers. Hidden layers do not interact with the outside world but help in non-linear feature extraction about data providing the output-input layer. | Data consisted of injection velocity, melt temperature, packing pressure, and mold temperature | NA | Parameters were optimized for minimum dimensional shrinkage. Reduces a good amount of cost and time. | Few results of reverse mapping deviation which is more significant than 10% | [117] |
|