1 Introduction

Tungsten carbide (WC) is a hard ceramic composite metal, with a polycrystalline microstructure, the binder used in sintering adds toughness properties to offset the brittle nature of the ceramic. It is the most widely utilised cutting tool material in a range of applications: turning, milling, high-speed cutting and high-value manufacturing [1]. The unique characteristic of having both metallic and ceramic properties gives WC high strength, durability and the ability to withstand high temperatures [2].

Cutting tools need to be manufactured to small tolerances to produce the desired geometry or features on a workpiece surface. Laser machining is a precise and non-contact operation for processing cutting tools [3], however, the process can induce both mechanical and thermal defects. In WC these microstructural defects are mainly cracking, thermal shock [4], porosity [5], recast layer and heat-affected zone (HAZ) [1]. These defects limit the tool performance and life [2] and act as stress raisers and points of crack initiation, as cutting tools are subjected to large loads and forces.

Cracks are narrow ruptures; the size and depth indicate the amount of damage caused. During laser processing, material phases undergo thermal expansion, expanding during the heating phase and contracting during the cooling/resolidification stage [6]. If there is a significant difference in the expansion rates, a thermal mismatch will occur leading to micro-cracks [7]. If this occurs in an uncontrolled manner across the microstructure, this leads to thermal shock, this is particularly of concern with composite materials like WC with a cobalt binder ([8, 9]). The characteristics of cracks are also defined by their specific location within a microstructure. In polycrystalline materials, these are commonly boundary cracks which run between element grains ([10, 11]), inter-granular cracks ([11, 12]) which run within material phases and radial cracks ([13]) which are located between the substructure of the processing area. Mao et al. [14] induced these types of cracks when changing the laser parameters to cut cBN-WC-10Co composites. At high-energy settings, there was a reduction in cracks but the processed material showed other thermal defects like HAZ.

Current methods in choosing the right parameters to minimise processing defects rely on human judgement or selectively testing possible settings and determining the right combination which gives the desired result. Marimuthu et al. [1] used a picosecond laser to machine WC; increasing the laser fluence between 1.27 and 17.3 J/cm2 increased the ablation depth and surface roughness of the achieved grooves. A simultaneous increase of frequency and speed created a better quality with no significant thermal defects, e.g. recast layer, micro-cracks and HAZ. The study also determined that the material removal rate was proportional to the average laser power. Similarly, Dumitru et al. [15] found that increasing the fluence from 0.4 to 2.8 J/cm2 caused a higher surface roughness with large pores in the processing crater and physical damage on the surface when processing various ultrahard materials with a femtosecond laser.

Recently, various statistical and analytical techniques, e.g. regression, orthogonal arrays, Taguchi design of experiments (DOE) and analysis of variance (ANOVA), are used to compare the effects parameters have on the operation output and statistically examine the significance of each parameter [16]. They are simple tools to understand multi-variate behaviour compared to finite element approaches which require high computational input and expertise. Ayyildiz et al. [17] used Taguchi experiment design for surface roughness optimisation in the drilling of fibreboard and ANOVA for experimental analysis to find the optimal roughness. The statistical approach also determined that the parameters followed a quadratic regression rather than a linear regression for roughness optimisation. Kara et al. [18] also demonstrated the use of Taguchi for grinding optimisation on AISI 5140 and compare linear and quadratic correlation to surface roughness with a high success. In laser processing, statistical optimisation has been conducted on hard and ultra-hard materials ([19,20,21]). For example, Wang et al. [22] used DOE to optimise a hybrid laser waterjet on polycrystalline cubic boron nitride (PcBN); laser power and waterjet pressure were the most significant factors on the cut quality. Altering these parameters improved the cut quality reducing taper size and HAZ thereby controlling phase transitions in the microstructure. Calderón Urbina et al. [23] conducted an analytical investigation into the material removal rate of laser ablation on WC. Laser power, burst-mode pulse overlaps and scanning pattern were the factors of interest. It concluded that the laser power and overlap were the most influential variables on the ablation process.

Previous research shows that even simple modelling provides an objective and reproducible methodology to understand processing and material behaviour. However, fewer works have reported the optimisation of laser parameters for brittle materials (i.e., WC) based on crack formation and surface integrity. In cutting tools, cracking is a type of surface defect that strongly hinders tool performance; therefore, optimising and controlling the laser process to avoid crack formation are needed as studies have done in other materials [24].

1.1 Scope of the paper

Laser manufacturing of WC requires a thorough understanding of WC behaviour to improve the quality of the process and ensure that the resultant microstructure does not have an adverse effect on the cutting tool performance. WC is inherently brittle and falls under the group of materials that will fracture under thermal effects. The cemented carbides’ response to mechanical and electrical processes, e.g. grinding, turning, electrical discharge machining (EDM) and electrochemical machining (ECM), has been greatly reviewed, but limited research is reported on WC response to laser beam processes, i.e. heating and melting. The objectives of the research paper are:

  • To experimentally investigate the effect of laser parameters on the surface morphology and integrity on a WC cutting tool material using a pulsed fibre laser (1060-nm wavelength, 70-W maximum output power). The parameters chosen were fluence, frequency, feed speed and hatch distance.

  • Generate a numerical linear model to predict the surface roughness in relation to the distribution of surface peaks generated by laser processing, using DOE optimisation techniques.

  • To investigate the crack formation and fracture behaviour of WC using ANOVA by developing a new method for classifying surface cracks produced from laser processing. This includes the analysis of surface crack generation based on the most influential factors to the surface roughness.

2 Methodology

2.1 Workpiece material

The study used a WC-Co12 % blank (Ultra-met Z57 corresponding to an industrial grade C5 ANSI standard, P50 ISO Grade) with a mean grain size of 4 μm and 1.60-mm thickness on a molybdenum carbide substrate. This grade is typically used in turning and milling on softer alloyed materials at medium speeds [2]. The chemical composition of the sample, measured by energy-dispersive spectroscopy (EDS), is given in Table 1.

Table 1 Chemical weight percentage composition of unprocessed WC-Co12 % disk
Full size table

2.2 Laser machining

A 70-W InnoLas micro-machining system (MMS) fibre laser 1060 nm was used, with a beam mode profile of M2 < 1.6; set up shown in Figure 1a. The laser is equipped with pre-programmed waveforms. Waveform 0 was used as a default setting: pulse duration of 240 ns, max pulse energy 1 mJ; this gives a maximum fluence of 0.20 J/cm2. This is lower than the WC ablation threshold of 0.45 J/cm2 in the ns regime [15]; thus, laser mechanism discussed in this paper focused on melting and partial ablation of the binder. The beam diameter at the focal height of 60.5 mm was 80 μm; the generated laser fluences were calculated based on previous research from Pacella et al. [25]. A 1-mm2 square was processed for each experiment using one pass in a path pattern as shown in Figure 1b.

Fig. 1
figure 1

a Experimental set up. b Laser raster pattern. c 2D average roughness parameters in the vertical and horizontal directions

Full size image

2.2.1 Experiment design

A 24 factorial design consisted of four factors (fluence, frequency, speed and hatch distance) with two levels, giving 16 experimental conditions, repeated three times. All other laser parameters and conditions were kept constant, including the number of scan repetitions and laser spot size. Table 2 and Table 3 outline the parameter settings for each experiment. A reference experiment was also conducted between the two levels of each factor. The low and high levels of hatch distances and feed speed give an overlap of 83% and 50% respectively, calculated based on previous research by Dahotore and Marimkar [26]. The selection of laser parameters was based on the settings available on the InnoLas MMS and common laser parameters used to process WC ([15, 27,28,29]).

Table 2 DOE experiment levels
Full size table
Table 3 Taguchi 24 orthogonal array factorial design—parameter settings for each experiment
Full size table

The DOE experiment also acted as a screening test to identify the main parameters affecting the microstructure and surface integrity, leading to further experiments, discussed in this section. The second set of experiment (Table 4) focused on understanding the material response to irradiance with fluence and frequency (v = 500 mm/s, d = 0.02 mm), with parameter values distributed between the minimum and maximum settings of the InnoLas MMS to highlight the micro-crack generation, crack density and microstructural characteristics. An orthogonal array of 49 experiments was made with 7 settings for both factors; it was repeated three times to increase robustness; using an orthogonal array reduced the number of experiments needed. ANOVA was applied to experimental results to identify the variation caused by changing process parameters on crack formation and crack density. It was also utilised to find the significance and impact of parameters on crack density.

Table 4 Experiment number for comparing fluence and frequency for understanding material crack generation (v = 500 mm/s, d = 0.02 mm)
Full size table

The feed speed was also investigated independently to understand other material phenomenon, with 6 experiments varied between 250 and 2500 mm/s (ψ = 0.050 J/cm2, f = 52.5 kHz, d = 0.02 mm), repeated three times (orthogonal array).

2.3 Characterisation

Post laser processing, the samples were cleaned with an acetone solution to remove dirt and debris prior to characterisation. The surface roughness and 3D profile were measured using an Alicona Infinite Focus White light interferometry (WLI) and analysed using MountainLab software. A lateral resolution of 3 μm and vertical resolution of 50 nm were utilised. The surface was levelled to remove tilt, and waviness removed with a Robust Gaussian Filter and 0.25-mm cut-off. Three roughness Ra measurements were taken vertically and horizontally for each sample and averaged. The vertical and horizontal surface roughness were separate to isolate specific details of the laser process. The horizontal roughness (RaH) captured detail on the surface along the laser path. The vertical roughness (RaV) captured detail through laser passes (Figure 1c). Various 3D roughness measurements were taken using Abbott-Firestone’s curve, but the reduced peak height (Spk) was the measure of concern also known as the peak distribution on the surface. A high Spk implies that the surface has many peaks. The initial contact with another surface will cause most of the forces to be confined to these small peaks, thus increasing the contact pressure. This leads to surfaces quickly eroding and being worn down, a necessary consideration in cutting tools. This variable is also useful to know for the amount of area for lubricant retention and debris collection available. Scanning electron microscope (SEM), TM3030 Hitachi backscatter SEM, was used to view the surface topography in particular to identify microstructure defects such as cracks, thermal shock, voids, pores, splatter and balling as a result of laser processing. These defects affect the performance of a material in application [30]. Energy-dispersive spectroscopy (EDS) analysis for chemical composition was also applied in various regions. The material removal was calculated based on previous research [25].

2.4 Crack analysis

A new method was developed to estimate crack density to characterise crack formation in laser treated specimens. Cracks were identified as regions of material separation distinctly different from other surface defects such as a splatter boundary that follows the molten material shape. Firstly, the cracks were classified into three categories based on size: superficial cracks, micro-cracks and deep cracks by SEM inspection of the samples. Superficial cracks were faint, hairline scratches (length up to 20 μm). Micro-cracks were thicker cracks or cracks longer than 20 μm (length up to 100 μm, width up to 3 μm). Deep cracks were cracks that caused significant damage/breakage or void-like failures (or length ≥ 20 μm and width > 3 μm). All the cracks in a 225 μm × 175 μm sample space were outlined, categorised and counted for each sample (Figure 2).

Fig. 2
figure 2

a Example of sample size for crack analysis, b examples of crack categorisation based on size and damage to surface, c representation of each type of crack from a sample

Full size image

The length and measurement criterion are based on the works of Denkena et al. [31] which studied micro-crack formation in hard metal milling tools. The length of cracks is measured by importing the SEM image into MATLAB and overlaying a straight line from one end to the other. Cracks which deviated from a linear shape were overlayed with several straight lines in segments. The ratio of the workspace compared to the actual SEM size was used to convert the length of the lines drawn to the micron scale. The quantity of each type of crack in the area was scaled up to estimate the crack density present on a 1-mm2 area, resulting in a crack density value for each type of crack. The crack characterisation extrapolation is a valid estimation as laser processing is well established for its repeatability and reliability [32]. The total number of cracks from each sample was then compared to the fluence and frequency settings used to process that sample, allowing for an ANOVA result to be computed. It was important to generate this technique to have a quantitative analysis of crack formation as there is currently no universal way of quantifying cracks based on type. As this is a visual technique, the sample space used will be dependent on the workpiece material. It should be a size that covers at least three laser paths with the magnification set to view surface cracks to a good resolution. The crack classification is only relevant to surface inspection; the penetration and depth of cracks are out of scope.

3 Results

Surface roughness is a clear way of characterising the surface quality after processing. The as-received sample had an RaH 0.4628 μm, RaV 0.5585 μm and Spk 0.9033 μm. Table 5 shows the roughness values from the experimental data using the corresponding levels described in the ‘Experiment design’ section. The levels were used to calculate a gradient value respective to each parameter and parameter interactions, calculated using the method described in [33] (Figure 3).

Table 5 Roughness parameter value for parameters and parameter interaction
Full size table
Fig. 3
figure 3

Gradient values of moving from each level for each parameter

Full size image

For all three roughness parameters, fluence had the greatest effect. An increase from 0.050 to 0.149 J/cm2 caused an increase of 0.1203 μm, 0.1644 μm and 0.4189 μm for RaV, RaH and Spk, respectively. For RaV, the interaction of frequency and speed was the next greatest effect. However, speed independently reduced the roughness, but frequency independently increased the roughness. The remaining effects are less contributory. RaH followed a different trend with frequency, the interaction of fluence and frequency and the interaction of fluence, frequency and feed speed cause the most increase in roughness value. Hatch distance had less of an effect suggesting that the energy accumulation from other parameters is more likely to cause a poorer surface than having a smaller pitch between the laser passes.

The factors that consistently reduced the roughness were speed, the interaction between fluence and speed, the interaction between frequency and speed and the interaction of fluence, frequency and feed speed; also confirmed with the lack of defects on the microstructure of certain samples. This suggested that increasing the speed improved 2D and 3D surface roughness, and when optimised with fluence and frequency, an even better result was achieved in terms of 2D and 3D roughness. Table 6 shows the percentage variation of Spk from the as-received surface using Equation (1). Experiments processed at a fluence of 0.050 J/cm2 had a ± 20% variation from the as-received.

Table 6 Percentage variation of Spk form the as-received sample compared to each experiment
Full size table
$$ Percentage\ variation\ \left(\%\right)=\left(\frac{\left( Spk\ value\ from\ test\right)-\left( As- received\ Spk\right)\ }{As- received\ Spk}\right)x100 $$
(1)

The distribution of the Spk values based on individual parameters is shown in Figure 4. The distribution of Spk values was very small when using the lower fluence compared to the higher fluence setting; the mean is 43% higher. Frequency followed a similar pattern where the lower setting had a much smaller distribution than the higher setting; and a 30 % increase in mean. The speed and hatch distance had a smaller difference, with the quartiles of the roughness values overlapping at the two levels. However, at a higher speed setting, the deviation was within a 10% variation of the as-received Spk value. Hatch distance had the least variational difference between the levels.

Fig. 4
figure 4

Box plot of Spk for a fluence, b frequency, c speed and d hatch distance

Full size image

Pareto analysis or cumulative frequency (Figure 5) was also conducted on the parameters and the interaction using absolute gradient values, to prioritise which parameters caused a greater change in roughness. Fluence had the highest contributory effect on all roughness variables; in other words, increasing the fluence is more likely to cause a higher surface roughness due to the more aggressive material processing irrespective of other laser parameters (see the ‘Discussion’ section). Therefore, it is a necessary factor to reduce the roughness and have a better surface homogeneity. In all roughness parameters, a variant of frequency and speed (or interaction) had the second greatest effect on the roughness, indicating that speed and frequency need to be viewed as linked parameters when considering surface roughness in both 2D and 3D parameters. Spk was affected by the interactive effects of parameters more where the interaction between fluence, speed and frequency and the interaction of frequency and speed had the next largest contribution to the Spk. This suggested the importance of considering parameters as related variables and not isolated.

Fig. 5
figure 5

Pareto effect of each roughness variable based on the parameters chosen. a RaV, vertical average roughness, b RaH, horizontal average roughness, c Spk, reduced peak height

Full size image

Figure 6 shows the main effects plot with the mean Spk value graphically; the magnitude of each gradient indicates how sensitive each factor was to the Spk; the main effects plot is a frequently used method for Taguchi analysis ([16, 34]). When fluence was changed from 0.050 to 0.099 J/cm2, Spk increased; the increase was more rapid from 0.099 J/cm2 to 0.149 J/cm2. A similar behaviour was observed for frequency with the increase from 35 to 52.5 kHz was greater than between 17.5 and 35 kHz. The effect of feed speed showed the opposite behaviour than fluence and frequency. As the feed speed increased from 500 to 1000 mm/s, there was a greater decrease in Spk compared to the very small decrease caused when the feed speed was increased from 1000 to 1500 mm/s. With the hatch distance, there is a decrease in Spk from 0.02 to 0.04 mm but a large increase from 0.04 to 0.06 mm.

Fig. 6
figure 6

Main effects of plot: a the effect of fluence on Spk, b the effect of frequency on Spk, c the effect of speed on Spk and d the effect of hatch distance on Spk. Larger absolute gradient greater change on surface roughness

Full size image

Experiment 15 had the highest average Spk value and variation of 4.156 μm and 400%, respectively. This is due to the energy transfer being very high in the areas that are directly exposed to the pulsed laser. These regions resulted in deep grooves on the surface with depth in the region of 10 μm. The larger hatch distance used in this experiment made the surface profile comparable to a textured surface with linear grooves (Figure 7), hence the very high Spk value.

Fig. 7
figure 7

a Roughness profile direction on experiment 15, b RaV 2D surface roughness profile for experiment 15 of 24 design

Full size image

3.1 Surface roughness modelling

Statistical analysis from the Taguchi method and empirical data aided the generation of coefficients used to create the linear model in Equation (2) with a minimised error percentage (1%):

$$ {Spk}_{Model}=0.9033+0.41885\psi +0.24755\psi f-0.21025\psi v+0.13255\psi d-0.15635 vd-0.30205\psi f v $$
(2)

where ψ is fluence (kJ/m2), f is frequency (MHz), v is speed (m/s) and d is hatch distance (mm).

The parameters included in the equation align with the experimental data investigated in this study. For example, a larger fluence increased the peak distribution as there was more aggressive material melting on the surface leading to more molten material distribution and peaks. The model also takes the interaction effect of parameters to show how they limit or contribute to peaks on the surface. Figure 8 shows this model against the average Spk results measured from experimental data.

Fig. 8
figure 8

Model compared against Spk results (with error bars) and percentage error

Full size image

The model prioritised fluence and the interaction effects between frequency and speed shown in the previous analysis. It predicted the roughness in the melting region of laser processing with an accuracy of 69%. Eduardo et al. [35] reported a similar accuracy when modelling the melted layer geometry from laser additive manufacturing in Inconel 625 and Iron. The Spk model performed best at moderate to medium level of processing (fluences in the region of 0.050–0.099 J/cm2) with an error between 1 and 10%. The material removal was approximately less than 0.27 μm3. When the fluence was 0.149 J/cm2, there was more violent material melting and expulsion; the material removal was higher than 0.40 μm3, approximately. The model failed to predict this behaviour accurately with the error percentage between 15 and 36%. This suggests that different models may be needed to fully model the transition from the melting phase to the boiling/vaporisation phase. This is not an uncommon result when studies have attempted to model laser processing. Volpe et al. [36] created a predictive model for the femtosecond laser micro-milling on acrylic (PMMA). The linear model described most of the results but required adaptation using regression was needed to cause certain settings to fall within a good confidence level at higher energy settings. Schatz and Patel [37] also found this with DOE modelling of pulsed laser deposition on lead zirconate titanate (PZT) thin films. Multiple models were generated but no one model perfectly fell within a good confidence level, but the study was able to conclude that power and pressure were the most influential factors on the PZT properties.

Fluence caused more melting on the surface and melt redistribution, which increased the chance of microstructural changes and poorer surface. As the cooling rate was not controlled, the solidification rate varied in different regions. Some areas experienced faster cooling leaving small grains and some areas slower cooling leaving larger grains. The mixture of these rate caused voids, porosity and transitional boundaries characterised by peaks on the surface. The trends shown with these parameters are in line with the current material science of laser processing. Davoren et al. [5] studied laser net-shaping WC-Ni9.2% with a fibre laser with powers ranging from 150 to 350 W. A higher power caused a larger molten pool, resulting in a greater change in the microstructure. This created areas of different morphologies of spheroids and fish-bone structures, close to the processed area. Xiong et al. [38] presented similar results when laser net-shaping WC-Co, where a higher power created a greater variation in WC particle size causing voids and porosity. There was also a larger HAZ in the surrounding region.

The results of the study and model concluded that fluence is the most influential factor on surface peak generation. The higher energy caused more aggressive material processing and redistribution. This highlighted that fluence is a necessary factor to be optimised to minimise the roughness and have better surface homogeneity overall. The interaction effects of parameters were also significant, indicating that the surface quality cannot solely be optimised using one parameter and parameters cannot be looked at independently. The interaction of fluence and frequency and the interaction of speed and frequency should be considered correlated. A faster speed reduced the number of thermal defects present. However, the frequency had a greater effect on crack formation and the types of cracks formed. The configuration of the linear model generated indicated which parameters contributed most to influencing the surface roughness. Under moderate to medium level of processing, the linear relationship was accurate in describing the resultant surface roughness. It is worth considering the use of quadratic or higher polynomial model for more aggressive processing. The model is significant in providing a better understanding of the laser parameters and the interaction of the laser parameters on the surface integrity of processed WC.

3.2 Surface crack optimisation

Cracks are detrimental to the surface integrity and mechanical properties of WC post laser processing ([6, 39]). In this paper, a micro-crack classification system was developed, and a crack density variable was introduced to estimate cracks and crack type on a 1-mm2 area size. Fluence and frequency settings, between 0.05 and 0.20 J/cm2 and 5 and 100 kHz, respectively (Table 4), were investigated to isolate these parameters against the crack formation, with a constant feed speed and hatch distance; fluence and frequency settings were investigated. Figure 9 shows the result of the crack density distribution and Figure 10 shows instances of each type of crack at various processing settings. The crack density increased as the fluence increased but peaks at 0.099 J/cm2 for most frequency settings. Deep cracks were signs of significant damage in the area; on average they did not present until 40 kHz and were situated around balling and splatter defects. Deep cracks were present mostly above 0.099 J/cm2 across the frequency settings. Micro-cracks were mainly present with the lower fluence settings across the range of frequencies, around the main laser track path area. They were more likely to form between 0.050 and 0.135 J/cm2. Superficial cracks were present in all frequency settings but particularly with lower fluence settings. This crack type was highest at 0.070 J/cm2 and 35 kHz. The crack density of each crack type increased between 0.050 and 0.099 J/cm2 across all frequency settings and then decreased as the fluence increased to 0.20J/cm2.

Fig. 9
figure 9

Crack density against frequency at the following fluence settings: a 0.050 J/cm2 , b 0.070 J/cm2 , c 0.099 J/cm2 , d 0.135 J/cm2 , e 0.149 J/cm2 , f 0.200 J/cm2 (speed 500 mm/s, hatch distance 0.02 mm)

Full size image
Fig. 10
figure 10

Crack classification against different fluence and frequency settings, diagrams with a green square mark did not present cracks of that type

Full size image

The transition to thermal shock fracturing can be seen at 0.099 J/cm2 where there was an equal distribution of the crack types. The lowest density of all three crack types occurred at 0.149 J/cm2 fluence and 52.5 kHz; the microstructure showed a good amount of melting, uniform material processing and resolidification (Figure 11a). Overall, these results suggested that there was a link between crack defect formation and the variation of fluence and frequency.

Fig. 11
figure 11

a Minimal cracks with uniform processing, b typical location of each type of crack identified

Full size image

The types of cracks identified on the samples loosely followed a pattern. Superficial cracks were randomly distributed throughout the sample, in no specific regions or areas. Micro-cracks mainly presented within the main laser track area, i.e. along the RaH profile. Deep cracks were found in the overlapping laser track regions. Figure 11b generally shows this crack distribution.

The cracking behaviour of the material was investigated using ANOVA using the crack density variable against fluence and frequency; only these parameters were chosen as they proved to be the first cause of crack formation within the microstructure (see the ‘Discussion’ section). Table 7 shows the results of this analysis based on fluence and frequency with the percentage influence, calculated using the method described in [40]. The sample size included 147 specimens with a significance level (p value) performed at 10%. The magnitude of the f-statistics paired with a low p value less than 0.1 indicated that fluence and frequency independently had a considerable effect on the crack density. Fluence had a greater contributory magnitude. However, the interaction of fluence and frequency was not a substantial effect in crack density unlike surface roughness. The outcome of this analysis showed that crack generation can be altered by specific laser parameters. Future work should investigate the refinement of the fluence setting to reduce crack formation and subsequently frequency to quantify a threshold based on WC composition and relative laser system setups. The results showed that there was sufficient evidence to support the claim that crack formation can be altered by fluence and frequency independently, but it is not dependent on the interaction of both. This is the first time cracks have been analysed with the statistical process in the laser processing of WC, providing an insight into understanding the key laser parameters on the crack formation mechanism.

Table 7 ANOVA results for crack density using fluence and frequency
Full size table

4 Discussion

The as-received microstructure had a highly homogeneous crystallinity (Figure 12). Experiments referenced in this section refer to Table 3. The processed samples showed different levels of discoloration; samples processed at 0.149 J/cm2 and 52.5 kHz had noticeable regions of darker regions indicative of burning. In these regions, there was a high percentage of carbon (Table 8), possibly due to the high energy causing material disassociation on the sample (Figure 13). This is an effect when WC is exposed to high temperatures, causing the carbon in the WC phases to diffuse into the binder phases [41]. For example, Chen et al. [42] found this in the laser additive manufacturing of WC-reinforced iron and Bartkowski et al. [43] noticed this effect when laser cladding Fe/WC on low carbon steel.

Fig. 12
figure 12

Microstructure of the as-received WC sample

Full size image
Table 8 Chemical composition at burnt regions within samples
Full size table
Fig. 13
figure 13

Burnt regions in samples

Full size image

Majority of the samples had a distinct HAZ around the 1-mm2 processing region, with a border of approximately 0.25 mm; this is expected as carbides often suffer from HAZ to different degrees depending on the processing parameters. Dumitru et al. [15] explained this is due to the small optical penetration depth of WC, which causes heat diffusion in the laser processing area. The microstructure of the laser-treated surfaces indicated that temperature was too low for the surface to properly melt all phases of the material [44, 45]. The binder and WC phase were thermally excited to different extents due to the difference in thermal properties within the material (Table 9 [46]). This resulted in the array of cracks throughout the surface by the expansion and contraction of the binder.

Table 9 Thermal properties of materials associated with Tungsten Carbide [46].
Full size table

4.1 Effect of fluence and speed on material processing

Fluences above 0.099 J/cm2 resulted in a significantly changed microstructure where there was a lot of molten material redistribution. This is caused by the photon momentum of the process [47]—causing a splatter/splotchy geometry; these features had a diameter of around 45 μm (Figure 14a).

Fig. 14
figure 14

a Surface remelting and splatter geometry from laser processing, b experiment 6, c experiment 7, d experiment 15, e experiment 16

Full size image

The ridges surrounding the circular geometries explained the increase of peaks and valleys on the surface. Analytically fluence had the greatest effect on peaks on the surface; this is also the case on a microstructural level. When comparing all the samples that were processed at 0.149 J/cm2, they had a resembling microstructure despite changing the other parameters. This indicated that there was a fluence threshold, between 0.099 J/cm2 and 0.149 J/cm2 on the WC structure, above which fluence dictated the resulting microstructure. All the experiments processed at 0.050 J/cm2 had greater variation in appearance and surface morphology suggesting other parameters take precedent on the resultant microstructure.

Experiments processed with 0.149 J/cm2, 52.5 kHz and 500 mm/s transferred the most amount of laser energy; they appeared burnt and had the most remelting in the region (Figure 14b, Figure 14d). This is as expected as the high fluence and frequency paired with the slower speed prolonged the amount of energy delivered—giving time for the workpiece to heat and melt but not ablate ([15, 48]). In some samples, the laser raster pattern was evidenced by material peaks and bulges of remelted material (Figure 14e). The effect of fluence on the surface topography is shown in Figure 15. The 0.149 J/cm2 fluence caused more melting, with material migration resolidified on the surface giving the splatter like appearance referred to before. There were also fewer cracks, transverse to the laser beam path. Marimuthu et al. [1] found a similar trend when using picosecond laser to process WC. As the fluence increased the material removal and crater roughness increased, however less cracking damage was present in the surrounding microstructure. The 0.050 J/cm2 samples exhibited smaller circular regions of melt expulsion and voids with a greater distribution of cracks.

Fig. 15
figure 15

a Experiment 9 and b experiment 12

Full size image

Increasing the speed showed a clear improvement in the quality of surfaces. Speed dictated how long each area of the workpiece was exposed to the laser and changed the amount of energy deposited. A slower speed increased the amount of thermal damage to the surface giving rise to a range of defects. The samples processed with the faster speed of 1500 mm/s had a more homogeneous surface in relation to uniformity and the distribution of defects. This was attributed to the laser energy causing localised heating and the momentum of the process distributing molten material causing splatter to spread through multiple laser tracks (Figure 16b). A slower speed concentrated more heat in specific areas, giving rise to even more splatter and localised defects (Figure 16a).

Fig. 16
figure 16

a Experiment 5, b experiment 11. White squares represent localised defects

Full size image

Speed also altered the general response of the material to the laser. Experiment 3 showed significant thermal shock evidenced by numerous superficial and micro-cracks (Figure 17). This was a result of the slow speed and higher frequency combination, as other samples processed with 500 mm/s and 52.5 kHz also presented a similar crack distribution. The WC grades used in this study had a relatively low WC content with a greater presence of the cobalt binder. As the WC and binder have very different thermal expansion coefficients, 5.5 μK-1 and 13 μK-1, respectively; each constituent expands at a different rate causing internal stresses and cracks. This was also demonstrated in the SLS of WC-Co by Paul et al. [6], where the rapid cooling of the cobalt phase caused high surface tensions in the molten pool. This caused cracks and geometric distortions. A higher weight percentage of WC would reduce this effect if the same parameters were used as less of the binder would be present to induce stresses [30]. Khmyrov et al. [39] showed this when laser sintering WC-Co of equal weight percentages and found many cracks due to the formation of W3Co3C brittle phases.

Fig. 17
figure 17

Thermal shock demonstrated in experiment 3

Full size image

A further look into speed on the material processing was done using the 0.050 J/cm2 to limit the effect of fluence on the surface (Figure 18).

Fig. 18
figure 18

Microstructure of workpiece when processed at different speeds: a 250 mm/s, b 500 mm/s, c 750 mm/s, d 1000 mm/s, e 1500 mm/s, f 2500 mm/s

Full size image

At slower speed, there is a greater build-up of heat allowed for greater energy transfer. This increased the likelihood of the thermal mismatch, which presented as cracks and voids particularly at the carbide/binder interface [12]. At 250 mm/s, cracks were concentrated and prevalent near voids (Figure 19). As the speed increased, there was more irregular heating causing porosity (Figure 19). At 750 mm/s, the voids became larger and more frequent on the surface (Figure 20). The fastest speed of 2500 mm/s exhibited regions of barely processed areas surrounded by porosity (Figure 21); from the non-uniform heating and rapid solidification, restricting the material’s ability to melt and remelt uniformly [49].

Fig. 19
figure 19

250 mm/s feed speed result

Full size image
Fig. 20
figure 20

750 mm/s feed speed result

Full size image
Fig. 21
figure 21

2500 mm/s feed speed result

Full size image

4.2 Effect of frequency on the fracture mechanism

Frequency affected the amount of energy the laser transferred to the surface; the higher the frequency the greater the potential heating effect. Previous analysis showed that frequency was a clear factor in crack formation on the surface [6]. As frequency had such a dramatic effect on the surface, a further inquiry into material response was conducted. Figure 22 shows the result using higher frequency levels. SEM analysis showed that thermal excitation at the high frequency changed the material response to crack generation.

Fig. 22
figure 22

Microstructure of workpiece when processed at different frequencies: a 5 kHz, b 17.5 kHz, c 28 kHz, d 40 kHz, e 52.5 kHz and f 100 kHz

Full size image

At the lowest frequency of 5 kHz, there was a heterogeneous energy distribution demonstrated by the clear regions of material transition, similar to when a large hatch distance is used (Figure 22a). The boundary laser regions and overlapping tracks was characterised by voids, balling, and irregular pores (Figure 23). At 17.5 kHz, the surface showed more mechanical effects of cracking; with time for molten material to distribute (Figure 22b). The thermal response from the laser, from 40 kHz, began to dominate the morphological changes, altering the characteristics of the cracks and defects balling became more pronounced and frequent (Figure 24); caused by Marangoni convections from the higher temperature gradient. This caused the material to solidify before spreading along the surface. This phenomenon has mainly been found in SLS. Zhou et al. [50] studied the balling of pure tungsten with SLS by changing the laser exposure time. A longer laser exposure time caused thermal convection currents in the molten pool making the balling more distinct and larger. This balling defect has also been present in other materials. Khairallah et al. [51] investigated the physics of laser powder bed fusion of stainless steel and found that the balling effect is a result of the surface tensions. The balling was minimised by controlling the heat concentration on the laser processed area. From 52.5 kHz, a greater heat accumulation in the surface caused more remelting and splatter like regions. The movement of splatter was also confined to the width of the laser track indicating insufficient time to distribute around the surface (Figure 22e). The 100-kHz frequency setting had a distinct surface morphology, with fewer defects of balling and spherical pores (Figure 25). It was evident from this that there was a progressive transition from mechanically induced stress from the laser to a thermally driven response. This trend has also been seen in the laser processing of gold alloy thin films by Geremia et al. [52].

Fig. 23
figure 23

5-kHz frequency result

Full size image
Fig. 24
figure 24

40-kHz frequency result

Full size image
Fig. 25
figure 25

100-kHz frequency result

Full size image

5 Conclusion

The presented findings show insights in understanding the surface integrity of laser surface engineered tungsten carbide. The material response to different laser process parameters (fluence, frequency, speed and hatch distance) was evaluated, using a pulsed fibre laser at 1060-nm wavelength in the ns regime pulse duration. The conclusions are summarised:

  • The study shows a need to understand the relationship between the laser parameters chosen for processing WC. Laser-induced damage on WC ranged from cracking, thermal shock, porosity, balling, HAZ, splatter and spherical pores. The parameters altered the amount of energy delivered to the workpiece material. Fluence was the most influential factor on the surface roughness and microstructure. It dictated the morphology of features ( > 0.149 J/cm2) in the processed regions in a splatter-like geometry, which in turn increased the surface roughness. Increasing the frequency (5–100 kHz) caused a progressive transition from mechanically induced stress to a thermally driven response, with more thermal defects notably balling. Feed speed had the greatest effect on defects such as voids and porosity. Slower speeds (250–750 mm/s) caused concentrated heated areas, giving rise to localised defects, e.g. pores, porosity, and increased the likelihood of thermal shock. At higher speeds (> 2000 mm/s), the microstructure exhibited regions of barely processed areas surrounded by porosity from non-uniform heating and rapid solidification, restricting the material’s ability to melt and remelt uniformly.

  • The crack classification system was useful for objective and quantitative analysis. It classified cracks into 3 groups: superficial, micro-cracks and deep cracks based on size and damage. A crack density variable was calculated to estimate the number of cracks and crack type within a 1-mm2 area size based on the laser parameters. Despite this, the process is still a rough approximation and is not automated, future research will investigate the use of computer processing techniques, e.g. pixel analysis, to improve the robustness of crack identification and classification.

  • For minimising the crack density, fluence (63.49 %) and frequency (29.38 %) had a significant effect on the crack density independently but not the interaction of both (7.13 %). The crack density increased between 0.050 and 0.099 J/cm2 across all frequency settings then decreased as the fluence increased to 0.20 J/cm2. Superficial cracks were present in all frequency settings but particularly with lower fluence settings. Micro-cracks were more likely to form between 0.050 and 0.135 J/cm2. Deep cracks did not present until 40 kHz or above 0.099 J/cm2 across the frequency settings and were situated around balling and splatter defects.

  • A linear model to predict surface roughness was proposed based on the experimental Taguchi analysis. It performed best when there was a moderate to medium level of processing (0.050–0.099 J/cm2), with an error between 1 and 10 % but failed to accurately model the behaviour when there was more violent material melting (> 0.099 J/cm2), with an error between 15 and 36 %. It can be concluded that the interaction of parameters is as important to optimise as the parameters themselves independently. This process showed the laser process can be modelled however different models may be required to describe the behaviour at different processing states, i.e. melting or ablation.

  • To build more confidence in the robustness of optimisation, more advanced modelling techniques will be considered including artificial intelligence (AI) and deep learning. These methods can handle a greater number of parameters and settings to analyse the surface integrity. AI processing can also aid in inspecting surface cracks with visual inspection techniques. The aim of study was to develop a method to understand and control the laser processing of brittle materials to retain the surface integrity for various applications.

  • The findings of the present work are important as they provide a basic framework for understanding the material science interaction during the laser processing of WC and how dependent the factors are on the microstructure of samples. In particular, cracks induced within the microstructure have an impact on tool performance in industrial applications, limiting the life. On the other hand, specific features, e.g. textures, generated on the microstructure may be advantageous to certain machining applications. The optimisation of laser parameters can minimise defects and control the generation of specific features.