A Review on Machine Learning Models in Injection Molding Machines

Table 9

Hybrid algorithms in injection molding.

Algorithm used	AI model taxonomy	Depth layer sizes, training time, and testing time	Dataset	Framework, core language, and interface	Advantages	Disadvantages	Ref.
Self-designed learning algorithm	Supervised learning	NA	The dataset consists of process control parameters	MATLAB toolbox	It provides real-time control. Effective in the controlled injection molding machine.	NA	[97]
A hybrid learning algorithm is used	Supervised learning	Two-layer network	Dataset consists of peak melt temperature, cushion, dosage time, cycle time, injection time, and ram velocity	NA	High process reliability. Cost-effective due to reduction in production costs.	Automatic tuning of machine setups	[98]
NA	Supervised learning	Three-layer network	Data from experimental investigations	Product design workbench	Quality of the model is very high for the parameter space investigated, allowing plausible formulation-property correlations to be readily presented	NA	[99]
An efficient algorithm, reliable in detecting the presence of the location of weld lines	Supervised learning	3-4-1 three-layer BPN is modeled for evaluating weld line properties	Experimental data for open literature simulates results about self-development (CAE)-system HSCAE	MATLAB neural network toolbox	Detecting the location about weld lines with good accuracy	NA	[100]
Multiple learning algorithms	Supervised learning	One or two hidden layers	Review of literature on injection molding, casting, and applications through ANN	Fluent, CALCOSOFT, and Magmasoft	Provides a comprehensive exploration of types of research performed on injection molding parameters to obtain the best result	NA	[101]
Bee colony algorithm	Semisupervised learning	Packing time is 6.6 s, and cooling time is 3.4 s	Dataset consists of packing pressure (Pack_P), mold temperature (Mold_T), and melt temperature (Melt_T),	FE analysis software Moldflow	Has advantages in terms of quality and costs and efficiently supports engineers to determine the optimal process parameter	Many design variables related to machine settings and mold conditions were not considered	[102]
Simple SOM model	Unsupervised learning	One or two-dimensional network	Data were melt temperature levels mold temperature levels, holding pressure levels, packing time, and cooling time	NA	Efficient method to find an optimal parameter for the injection molding process	The amount of needed data is large	[103]
Particle swarm optimization algorithm	Supervised learning	A factor with a significant impact is time for cooling. It tells about shrinkage about HDPE.	Data consist of input parameters: mold temperature, packing pressure, packing time, and melting temperature	NA	Results accurately show melting of the temperature of 190°C, injective pressures about 70 MPa, and refill pressures of 85 MPa, and minimum shrinkage is found within eleven seconds of cooling	NA	[34]
Multiple response optimization models	NA	3-layer network, where input layer has 5 nodes and output layer has 3 nodes	Data consist of 5 process parameters: values cooling time, mold temperature, the melt, the injection flow rate, and holding pressure	MATLAB software	Improves the amount about proposing multiple responses and optimizes the problem about multiple responses optimized in the enterprise	NA	[104]
Neural network model, an integrated simulation program	NA	Three-layer adductive network	Data consist of the length and diameter of the runner system	FEM simulation	Rapidly and efficiently gives the result for determining the optimal runner-system parameter about injection molding	NA	[12]
Levenberg–Marquardt algorithm	Supervised learning	Multilayer FFN	Data consist of temperature in the melting zone, the pressure of molten plastic, and mold tolerance	MATLAB/Simulink software environment	Significantly less overshoot and faster transience	Has a larger quantity of neurons	[105]
DFSMC algorithm	NA	Frequency fs = 200 Hz, oil temperature toil = 30 ∼ 40°, and maximum supply pressure pS max = 21 MPa.	The total control system is coupled with 2-input 2-output system	NA	Counteracts a coupling effect, reducing a fuzzy rule number	NA	[106]
Fuzzing sets and neural networks	Supervised learning	Input and output layers have 45 and 81 neurons, respectively.	Dataset consists of short-shot, sink-mark, ash, ow-mark, weld line, warpage, and cracking	MATLAB neural network toolbox was used for system implementation	Predicts the parameters perfectly at the test run itself without wasting many resources	The neural network should be trained with exceptional white rules to encounter the sudden unexpected problems	[107]