|
| Algorithm used | AI model taxonomy | Depth layer sizes, training time, and testing time | Dataset | Framework, core language, and interface | Advantages | Disadvantages | Ref. |
|
| An artificial neural network trained with a backpropagation algorithm | Supervised learning | It consists of a two-layer network, and cycle time is 43.77 seconds | Dataset consists of 6 quality variables (part length, spot, stains, burn marks, flash, and unfilled parts) | NA | All the parameters have high coefficients, which shows that the network performs well | There is a small amount of redundancy among all the observation variables | [79] |
| ANN coupled with an improved PSO algorithm like hybrid ANN-PSO | Supervised learning | Multilayer feed-forward backpropagation ANN | The dataset consists of forty-four sets, where forty sets are utilized as the training set, and four sets are utilized as test set | Python | There is an excellent agreement between measured radii about curvature about a bi-aspheric lens with less than 1% deviation and the predicted ones | There are strain and volumetric shrinkage variations in the lens | [80] |
| A backpropagation neural network algorithm | Supervised learning | It consists of input and output layers | Test data as 2 models, respectively, consisting of 120 samples and 40 samples | NA | Increasing the predicted performance on a product quality | NA | [81] |
| A simulated annealing (SA) optimization algorithm | Supervised algorithm | Three-layer abductive networks | The datasets consisted of various runner diameters, cavities, volumes, and gate lengths, searching for max warp and gate diameter. | Moldflow/MPI system | Accurately predicts the warp of multiinjection mold | There is an error of 8.6%, valuing optimally selective values about results in prediction of neural models | [14] |
| Backpropagation neural network algorithm | Supervised learning | Two layers | Data consisted of input conditions: CaCo3%, MnO3%, dwell time, and temperature | MATLAB 7.0 | Predicts the output range which is lesser than ten percentage errors that correspond to the noticed value | NA | [82] |
| Artificial neural networks, trained with the backpropagation algorithm | Supervised learning | Three-layer network | Data were obtained from 126 production cycles | QwikNet | Even with less important variables than desired, an ANN system to diagnose faults can be successfully built | Only one topology was considered | [13] |
| Multilayer perceptron (MLP) backpropagation algorithm | Supervised learning | It consists of three layers | Dataset consists of injection speed and holding pressure | Integrated circuit (IC) | Reduces the size of data, which improves the training time | Prediction of the quality is good if more datasets are used | [26] |
| Non-binary genetic algorithm | Supervised learning | Three-layer neural network with a 4-12-1-neuron configuration | There are 54 training datasets and 10 testing datasets | Simulation studies were done utilizing Moldflow software systems. | The process operating parameters were determined correctly | The training dataset size should be increased | [83] |
| Error backpropagation algorithm and the Levenberg ± Marquardt approximation algorithm | Supervised learning | Two-stage multilayered feed-forward neural network | 114 total data points in which 94 are used to train | C-MOLD simulation software and MATLAB | Have the ability to predict injection time while presenting new operating condition with high accuracy | Slow convergence and a tendency about the networks from becoming struck locally, making a gradient in a descent methodology not suitable or applicable for applications of this type | [84] |
| Backpropagation artificial neural network algorithm | Supervised learning | The 2-hidden-layer architecture was utilized for every network | Dataset consisted of process parameters | Software simulation. Using a PC with spreadsheet input. | Able to process and train in time using the single network just for monitoring multiple products, qualities, and parameters | Case study of real-life manufacture problems was not considered | [24] |
| Error backpropagation algorithm and Levenberg–Marquardt algorithm | Supervised learning | 3-layer network | Dataset consists of process parameters | C-MOLD simulation and MATLAB | Overall error is significantly less than 0.93% | NA | [17] |
| Backpropagation (BP) algorithm | Supervised learning | Two-layer network | Data consist of process parameters | PLASFIL | Increase of accuracy of a process | The number of process parameters was less | [85] |
| Scale conjugating gradients and Levenberg–Marquardt (LM) learning algorithm | Supervised learning | Three-layer neural network | Some experimental data were utilized for testing data; that value is not utilized as train.
MAPE and R2 for testing data were 1.849 and 0.9995 | TOOL-TEMP TT-157 E MATLAB | Predicts the yield lengths of plastic in the mold | NA | [86] |
| Backpropagation learning algorithm | Supervised learning | Three or more layers | Dataset consists of mold, constant pressure, melt, gas injection, injection speed, gas delay, and gas distance. | NETS simulator | Improved through reducing residual stress, warpage, shrinkage, and sink mark | Limited testing projects were used | [42] |
| Backpropagation algorithm | Supervised learning | Three layers | The data consist of powder blending parameters and are randomly split into test set and train set | MATLAB | Found out optimal solids loading of feedstocks consisting of blends | Not able to detect a critical solid loading enabling the attainment of a significant reduction in lead times for a particular feedstock | [37] |
| Backpropagation algorithm | Supervised learning | Three-layer network | The data consist of coolant flow rate, coolant flow temperature, and CTP | National Instruments simultaneous DAQ boards are used in combination with LabWindows CVI software | Excellent prediction about polymeric surface cavities’ temperature profiles based upon different process conditions | ANN size is not optimized | [87] |
| Backpropagation algorithm | Supervised learning | Three-layer network. The hidden layer is considered with 2 nodes. A value of 0.9 was utilized to reduce overfitting. | Data consist of MFR, Pinj, Tmold, and Tmelt | PartAdviser CAE | The model developed can predict standard injected part as a shortage of otherwise known about weld line defects or short shots | Process parameters are not integrated | [88] |
| Backpropagation algorithm | Supervised learning | 3-layer feed-forward backpropagation with ten neurons in the hidden layer | Data are extracted from the factory databases utilizing MATLAB | MATLAB environment | Reported through applying the intelligent system technique for improving downstream performances predicting in range of manufacturing environment | Several measurements are inadequate | [89] |
| Backpropagation algorithm | Supervised learning | Three-layer networks, where the hidden layers had 3 neurons, and the neurons are L-neuron, D-neuron, and P-neuron, respectively | Dataset consists of temperature | Visual Basic software platform | Train timing about the PIDNN was relatively small, and the last controlled results are excellent | It was computationally costly and time-consuming to train with traditional CPUs | [41] |
| Backpropagation-learning algorithm | Supervised learning | Three-layer network | Data are collected through rotation viscometry continued through the non-linear regression | NA | Reducing a quantity for the experiment needed for determination for practicing constantly | NA | [90] |
| Backpropagation-learning algorithm | Supervised learning | Three-layer network | Product cost data | MATLAB | Complete info in the process timing was no requirement. The neural networks help us improve concept design by evaluating costs on the various process and design alternatives. | Determines the total number as hidden layers; the neuron on every hiding layer was the trial-error process. The selected neural network might be the best and may not be the best. Sensitive studies about a cost relating feature in a product the cost might be challenging. | [91] |
| Backpropagation algorithm | Supervised learning | Three-layer network | Dataset is split into a training set, validation set, and testing set | MATLAB tool | Able to model almost any data | Relies on discrete measured process parameters as inputs to the model | [92] |
| Backpropagation algorithm | Supervised learning | Entry layer with five process variables, a hidden layer with 33 neurons, and an exit layer with eight variables. | Dataset consists of process variables like injection, temperature, and switch over. | Specific software not mentioned | The precision of the model of a neural network is perfectly valid | Very few process variables are considered | [93] |
| Backpropagation algorithm | Supervised learning | 4-2-3 (4 neurons of an input layer, two neurons of the hidden layer, three neurons of output layer) | Data were obtained experimentally based on the Taguchi test schedule | C-MOLD software | Reduces the number of experiments and has adaptive learning ability | Requires a lot of training process before a BPANN can be used for prediction and optimization | [94] |
| Backpropagation algorithm | Supervised learning | Three-layer network | Powdered metals, compaction pressure, powder composition, and sintering condition are important input parameters | MATLAB neural network toolbox | Demonstrated method is about advising the best selection of material as earlier as possible onstage about design in a component | Design geometries about commonly occurring component were not considered | [95] |
| Backpropagation algorithm | Supervised algorithm | The network had 3 core layers and output layers, a hidden layer, and an input layer | Data consist of pattern lines (two hundred micrometer nozzle), feed rates, pressure, and standoff distance (millimeter) | MATLAB simulation model | The approach adopted in this paper is applicable for different printable materials having a different composition and viscosity along with another printable technology about providing the incentive about optimizing | Dispensement in materials affects the volume of air in syringe barrels and the viscosity of the material | [96] |
| Backpropagation algorithm | Supervised learning | Three-layer backpropagation neural network | Data consist of surface finish, number type lines, part size, cavity material, tolerance, number of cavities, plating, mold material, number of cavities per mold, insert, and mold base type | CAD modeling software | Helps in increasing the efficiency of the product developing process by reducing the number of mold design iteration designed by the designer | Many input parameters are considered and it is sensitive to noisy data | [1] |
| Feed-forward type neural network | Supervised learning | 3-layer network, and the hidden layer was determined through the trial and error method | Injection time data were utilized for training a neural network system | MATLAB environment | Solves the real-world problem of the MIM process and Moldflow software | NA | [23] |
|
