Laser surface hardening is a very promising hardening process for ferrous alloys where transformations occur during cooling after laser heating in the solid state. The characteristics of the hardened surface depend on the physicochemical properties of the material as well as the heating system parameters. To exploit the benefits presented by the laser hardening process, it is necessary to develop an integrated strategy to control the process parameters in order to produce desired hardened surface attributes without being forced to use the traditional and fastidious trial and error procedures. This study presents a comprehensive modelling approach for predicting the hardened surface physical and geometrical attributes. The laser surface transformation hardening of cylindrical AISI 4340 steel workpieces is modeled using the conventional regression equation method as well as artificial neural network method. The process parameters included in the study are laser power, beam scanning speed, and the workpiece rotational speed. The upper and the lower limits for each parameter are chosen considering the start of the transformation hardening and the maximum hardened zone without surface melting. The resulting models are able to predict the depths representing the maximum hardness zone, the hardness drop zone, and the overheated zone without martensite transformation. Because of its ability to model highly nonlinear problems, the ANN based model presents the best modelling results and can predict the hardness profile with good accuracy.
AISI 4340 is a nickel-chromium-molybdenum alloy steel known for its toughness and its ability to attain high strengths when heat-treated, while retaining good fatigue strength. Typical applications are for structural use, such as aircraft landing gear, power transmission gears and shafts and other structural parts. This alloy may be heat treated to high strength levels while maintaining good toughness, wear resistance and fatigue strength levels, as well as good atmospheric corrosion resistance [
When compared with conventional hardening methods such as oven, flame, or induction hardening, laser hardening is marked by a range of advantages such as spatially and temporally limited energy deposition, which eliminates the need for quenching with water, oil, or salt baths. Laser hardening allows for a highly defined zone of influence without affecting neighboring surfaces, and high cooling rates make fine structures and high levels of hardness possible. Intricate contours are easily hardened using lasers due to the flexible beam guidance possibilities, also making it possible to harden parts directly where it is required [
In surface hardening using laser radiation, carbonic steel is heated above the austenization temperature for a short time. Through rapid cooling the steel reaches the martensitic material structure. Heat deposition is realized through the absorption of laser radiation at the surface of the material, whereas cooling occurs conductively within the remaining material [
There are many methods for modeling the laser hardening process, such as the finite element method (FE), the regression method, and the artificial neural network method (ANN). The principle of the FE method is solving the heat transfer differential equation to determine the evolution of the studied workpiece temperature versus time; the maximum temperature and cooling rate on the surface of the workpiece are calculated to determine the hardened zone. Many researchers used this method for plate [
The laser heat treatment of cylindrical pieces is done with a combination of two movements: a rectilinear movement of the articulated robot carrying the laser fiber, and a rotation of the cylindrical part generated by a DC motor. This combination of two speeds during treatment makes the finite element model simulation slower and difficult to calibrate, which leads us to choose ANN modeling methods. This is a powerful modeling tool that has the ability to identify complex relationships from data for nonlinear problems compared to regression modeling, despite the significant number of experimental tests required for the modeling.
Research based on ANN modeling for laser hardening on cylindrical workpieces is not numerous. In the literature there is one conference paper [
The experimentations are performed by using a commercial 3 kW Nd:Yag laser source (IPG YLS-3000-ST2) combined with a 6 degrees of freedom articulated robot. This type of laser generates a beam with a wavelength λ = 1064 µm. The AISI 4340 steel cylindrical parts, with 18 mm diameter 50 mm length, are mounted on a lathe as illustrated in
A rotating testing rig is used during this experiment to rotate the workpieces throughout the laser heat treatment, and the laser beam is set to point the median plane of the cylinder, and remain in it during the translation motion of the articulated robot. The cooling of the workpieces is done by radiation and conduction without any external cooling source. After the laser transformation hardening, the specimens were cut perpendicularly to the cylinder’s axis of revolution, and the required surface of each specimen was ground and polished with various grades of emery sheets. The hardened zone characteristics are determined by the surface hardness profile. Hardness measurements were done from the surface to the center of the workpiece, using Vickers’s tests with a load of 300 g applied for 15 s. For each trial, different hardness profiles with 50 μm steps were measured.
The process parameters in this article are the laser power (LP), the scanning speed (SS) of the laser beam, and the revolution speed (RS) of the part. Preliminary tests allow
the determination of process parameter limits so the non-hardening or melting of the workpieces can be avoided. Ranges of 1800 W to 2400 W laser power (LP), and 3 mm/s to 6 mm/s scanning speed (SS) allow the heat treatment of parts without producing melting or hardening discontinuity. The revolution speed is limited on the upper end by the limits of the DC motor rotating the workpieces, and on the lower end by the regularity of the hardened zone, so a range of 3000 rpm to 6000 rpm is used in the test matrix.
Due to its simplicity and transparency, a systematic design based on the factorial design is the most commonly used model in experimentation. However, this requires a substantial number of tests depending on the number of factors and their levels. In this way, the use of a testing strategy such as the orthogonal arrays developed by Taguchi leads to an efficient and robust fractional factorial design of experiments that can collect all the statistically significant data with a minimum number of repetitions [
An L9 Taguchi test matrix composed of central tests of L16 Taguchi matrix is used to validate the model is presented in
Test | LP (W) | SS (mm/s) | RS (rpm) | Test | LP (W) | SS (mm/s) | RS (rpm) |
---|---|---|---|---|---|---|---|
M1 | 2400 | 3 | 3000 | M9 | 2000 | 3 | 5000 |
M2 | 2400 | 4 | 4000 | M10 | 2000 | 4 | 6000 |
M3 | 2400 | 5 | 5000 | M11 | 2000 | 5 | 3000 |
M4 | 2400 | 6 | 6000 | M12 | 2000 | 6 | 4000 |
M5 | 2200 | 3 | 4000 | M13 | 1800 | 3 | 6000 |
M6 | 2200 | 4 | 3000 | M14 | 1800 | 4 | 5000 |
M7 | 2200 | 5 | 6000 | M15 | 1800 | 5 | 4000 |
M8 | 2200 | 6 | 5000 | M16 | 1800 | 6 | 3000 |
Test | LP (W) | SS(mm/s) | RS (rpm) | Test | LP (W) | SS (mm/s) | RS (rpm) |
---|---|---|---|---|---|---|---|
V1 | 1900 | 3.5 | 4500 | V6 | 2100 | 5.5 | 3500 |
V2 | 1900 | 4.5 | 3500 | V7 | 2300 | 3.5 | 3500 |
V3 | 1900 | 5.5 | 5500 | V8 | 2300 | 4.5 | 5500 |
V4 | 2100 | 3.5 | 5500 | V9 | 2300 | 5.5 | 4500 |
V5 | 2100 | 4.5 | 4500 |
The hardness curve obtained by laser heat treatment is illustrated in
The effects of different laser hardening parameters are determined using the hardness curve, which is characterized by identifying boundary points for the four different areas: D1 and H1 for the zone1/zone2 limit, D2 and H2 for the zone2/zone3 limit, and D3 and H3 for the zone3/zone4 limit. Throughout the article D1, D2 and D3 are expressed in µm, and H1 and H2 are expressed in HRC. H3 represents the initial hardness of the workpieces, and is fixed at 44HRC to determine D3.
The repeatability test is made to assess the robustness and reproducibility of the process, as well as errors resulting from measurement steps (50 µm between every two measurements of the microhardness) and shift of the measurement position. The central test of the validation table (2100 W for LP, 4.5 mm/s for SS, and 4500 rpm for RS) is repeated 10 times.
Test | D1 | D2 | D3 | H1 | H2 |
---|---|---|---|---|---|
R1 | 163 | 242 | 867 | 54.9 | 35.9 |
R2 | 158 | 237 | 817 | 55.8 | 36.2 |
R3 | 159 | 238 | 826 | 55.5 | 36.5 |
R4 | 153 | 230 | 845 | 56.0 | 36.7 |
R5 | 151 | 231 | 791 | 55.1 | 36.1 |
R6 | 159 | 240 | 841 | 56.4 | 36.3 |
R7 | 163 | 242 | 841 | 54.9 | 35.9 |
R8 | 159 | 240 | 836 | 55.5 | 35.5 |
Average | 157.8 | 238.3 | 835.8 | 55.56 | 36.04 |
S.D. | 3.8 | 4.4 | 21.3 | 0.48 | 0.40 |
Gap | 11.3 | 12.6 | 75.5 | 1.47 | 1.25 |
The gap between the maximum and the minimum hardness value of all the modeling and the validation data is 4.15HRC for H1 and 2.28HRC for H2, while the gap representing the repeatability and the measurement errors determined by the repeatability study is 1.47HRC for H1 and 1.25HRC for H2, which represents an error of 35% for H1 and 55% for H2. This is caused by the fact that the cooling rate, which is not controlled in the present work, is highly affecting the hardening process. On the other hand the gap between the maximum and the minimum depths values of the modeling and the validation data is 382.7 µm for D1, 490.3 µm for D2, and 1441.3 µm for D3, while the gap representing the repeatability and the measurement errors determined by the repeatability study is 11.3 µm for D1, 12.6 µm for D2, and 75.5 µm for D3, which represents an error of 2.9% for D1, 2.6% for D2, and 5.2% for D3. The error is low because these depths depend on the maximum temperature reached by heating and not the cooling rate [
The analysis of variance aims to study the effects of different parameters on the hardness profile. It gives the contribution of each parameter in the variation of process parameters.
Test | D1 | D2 | D3 | H1 | H2 | Test | D1 | D2 | D3 | H1 | H2 |
---|---|---|---|---|---|---|---|---|---|---|---|
M1 | 710 | 890 | 2561 | 56.2 | 35.3 | M14 | 105 | 148 | 800 | 52.4 | 35.6 |
M2 | 355 | 442 | 1325 | 56.3 | 36.5 | M15 | 56 | 120 | 521 | 53.7 | 36.4 |
M3 | 233 | 309 | 946 | 56.2 | 36.2 | M16 | 55 | 91 | 321 | 54.2 | 38.1 |
M4 | 116 | 194 | 736 | 55.3 | 35.7 | V1 | 200 | 278 | 1138 | 56.0 | 35.2 |
M5 | 532 | 698 | 2185 | 55.0 | 35.3 | V2 | 104 | 164 | 657 | 56.1 | 37.1 |
M6 | 278 | 364 | 1085 | 55.7 | 36.0 | V3 | 57 | 100 | 490 | 55.5 | 36.3 |
M7 | 163 | 242 | 854 | 56.5 | 36.6 | V4 | 347 | 442 | 1398 | 55.8 | 35.7 |
M8 | 105 | 145 | 639 | 53.4 | 36.6 | V5 | 159 | 240 | 836 | 55.5 | 35.5 |
M9 | 452 | 545 | 1819 | 55.9 | 35.8 | V6 | 105 | 155 | 574 | 56.0 | 37.2 |
M10 | 200 | 270 | 996 | 56.4 | 35.9 | V7 | 425 | 539 | 1532 | 56.5 | 35.4 |
M11 | 106 | 166 | 574 | 55.7 | 37.7 | V8 | 246 | 336 | 1105 | 56.1 | 35.4 |
M12 | 58 | 135 | 507 | 54.1 | 37.4 | V9 | 137 | 194 | 778 | 54.5 | 35.6 |
M13 | 238 | 345 | 1378 | 54.7 | 35.2 |
Source | DF | Contribution | Adj SS | Adj MS | F-Value | P-Value | |
---|---|---|---|---|---|---|---|
D1 | LP (W) | 3 | 22.73% | 123717 | 41239 | 11.74 | 0.006 |
SS (mm/s) | 3 | 68.84% | 374607 | 124869 | 35.56 | 0.000 | |
RS (rpm) | 3 | 4.56% | 24811 | 8270 | 2.36 | 0.171 | |
Error | 6 | 3.87% | 21068 | 3511 | - | - | |
Total | 15 | 100.00% | - | - | - | - | |
D2 | LP (W) | 3 | 22.55% | 173929 | 57976 | 13.04 | 0.005 |
SS (mm/s) | 3 | 69.58% | 536792 | 178931 | 40.25 | 0.000 | |
RS (rpm) | 3 | 4.42% | 34066 | 11355 | 2.55 | 0.151 | |
Error | 6 | 3.46% | 26675 | 4446 | - | - | |
Total | 15 | 100.00% | - | - | - | - | |
D3 | LP (W) | 3 | 14.97% | 905555 | 301852 | 10.53 | 0.008 |
SS (mm/s) | 3 | 81.21% | 4913854 | 1637951 | 57.17 | 0.000 | |
RS (rpm) | 3 | 0.98% | 59290 | 19763 | 0.69 | 0.591 | |
Error | 6 | 2.84% | 171915 | 28652 | - | - | |
Total | 15 | 100.00% | - | - | - | - |
(V1 − V9). The empirical relationship between the depths (D1, D2, and D3) and the laser process parameters can be expressed by Equations (1)-(3), which presents a model based on fit regression modeling and allows the evaluation of the depths (D1, D2, and D3) as a function of the laser parameters (LP, SS, and RS).
ANNs are relatively new computational tools that have found extensive utilization in solving many complex problems. The attractiveness of ANNs comes from their
remarkable information processing characteristics, pertinent mainly to nonlinearity, fault and noise tolerance, and learning and generalization capabilities. The idea of ANNs is not to replicate the operation of the biological system, but to make use of what is known about the functionality of biological networks for solving complex problems. ANN models are empirical, however, they can provide practically accurate solutions for precisely or imprecisely formulated problems and for phenomena recorded as field observations. These models learn from the preceding data obtained, which is named the training set, and then check the system accomplishment using test data [
The basic units of ANNs are neurons, which are connected to each other with a weight factor. ANNs are networks of highly interconnected neural computing elements that have the ability to respond to input stimuli and to learn to adapt to the environment. While a single-layer network has single input or output units, a multilayer network has one or more hiding units between the input and output layers. The multilayer perceptron consists of a system of simple interconnected neurons, or nodes, as illustrated in
The choice of the best ANN structure for modeling a given process is done by varying the number of hidden layers and the number of neurons in each one. Each ANN structure is trained several times to avoid the effect of the randomly chosen weight coefficients in the beginning of the learning phase. The first trial of ANN modeling was done with one hidden layer, and the number of neurons in the hidden layer was varied. Results show that the learning becomes perfect for over 20 neurons, but the model cannot be validated with the validation data. For less than 20 neurons the model has a high gap between prediction and experimentation, therefore the use of several hidden layers is beneficial. The number of neurons in each hidden layer is varied from 1 to 20, and the best model is found to have a neural structure of 3 neurons in the input layer (LP, SS, and RS), three hidden layers having the structure 2-15-8, and an output layer of 3 neurons (D1, D2, and D3). This model has a coefficient of determination, R2, of 1.000 for the modeling data and 0.995 for the validation data. The average error for all the modeling and validation data is 3.5% for D1, 2.2% for D2, and 2.5% for D3; the maximum relative error is 12.6% for D1, 13.1% for D2, and 10.7% for D3; and the gap between the maximum error and the minimum error is 46.7 µm for D1, 42.8 µm for D2, and 150.8 µm for D3.
means that the model is perfectly accurate compared to the regression model.
Using the modeling and validation tests, the fitness of the ANN model and the regression model (R-model) are characterised in
The ANN based model is much better than the regression based model, as shown in
In this paper, a predictive modeling based on regression and artificial neural network analysis of laser transformation hardening for cylindrical of AISI 4340 steel workpieces is presented. Several laser surface transformation hardening parameters and conditions are analyzed and their correlation with multiple physical and geometrical attributes is examined using a structured experimental and numerical modelling investigations under consistent practical process conditions. After identifying the parameters that provide the best information about the laser heating and the surface transformation process, conventional regression and artificial neural network analysis are proposed to build an accurate and consistent hardness profile prediction model. The process parameters included in the study are laser power, beam scanning speed, and the workpieces rotational speed. The upper and the lower limits for each parameter are chosen considering the start of the trans-formation hardening and the maximum hardened zone without surface melting (1800 W to 2400 W for the laser power, 3 mm/s to 6 mm/s for the scanning speed, and 3000 rpm to 6000 rpm for the workpiece rotational speed). The
Model | R squared | Average error | Gap (µm) | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|
RM2 | RV2 | RD12 | RD22 | RD32 | D1 (%) | D2 (%) | D3 (%) | D1 | D2 | D3 | |
R-model | 0.990 | 0.986 | 0.983 | 0.983 | 0.973 | 10.5 | 9.1 | 7.9 | 44.7 | 65.9 | 247.9 |
ANN | 1.000 | 0.995 | 0.997 | 0.998 | 0.995 | 3.5 | 2.2 | 2.5 | 46,7 | 42,8 | 150,8 |
hardened surface physical and geometrical attributes are extracted from the experimental microhardness curves. The results demonstrate that the resulting models are able to predict the depths representing the maximum hardness zone, the hardness drop zone and the overheated zone without martensite transformation. Because of its ability to model highly nonlinear problems, the artificial neural network based predictive model presents the best results and can predict the hardness profile with good accuracy. Globally, the performance of the hardness profile prediction model shows significant improvement. With a global maximum relative error less than 5% under various conditions, the prediction model can be considered efficient and have led to conclusive results, due to the complexity of the heat treatment process. The performance of the modelling process can be improved by including additional surface transformation hardening parameters and conditions. The inclusion of parameters like surface roughness and cooling rate in the predictive modelling procedures with others applications for complex geometric features are among the main directions to consider for the future works.
Jerniti, A.G., El Ouafi, A. and Barka, N. (2016) A Predictive Modeling Based on Regression and Artificial Neural Network Analysis of Laser Transformation Hardening for Cylindrical Steel Workpieces. Journal of Surface Engineered Materials and Advanced Technology, 6, 149-163. http://dx.doi.org/10.4236/jsemat.2016.64014