As for any other welding process, an unsuitable selection of process param- eters leads to unacceptable weld defects. Without effective monitoring and control strategy, many engineering solutions have to be considered in order to overcome quality related problems which will increase time and cost of production (You et al., 2014; Stavridis et al., 2017).
Traditionally, weld quality is assessed manually which consists of four steps:
(i) establishing satisfactory welding parameters by procedure trials and testing,
(ii) selecting and maintaining the same parameters/procedure in production, (iii)
monitoring process by final inspection (i.e. non-destructive/destructive testing to ensure that the required results are being achieved), and(iv)correcting for deviation from stated quality requirement by adjusting welding parameters. Furthermore, the final inspection may involve selecting random samples from a batch of finished welds and corrective action is normally based on statistical quality control techniques. The described approach, known asoff-line inspection, is costly, reduces productivity, and requires dedicated test equipment and people.
To ensure the acceptable weld quality, to increase productivity and to elim- inate the need for post weld examination, the welding process can be monitored during the process with sensors, known asin-process monitoring. The information gathered from sensors is then transmitted to the process controller that fine adjusts the welding process parameters to produce consistently acceptable quality welds. It must be pointed out that the weld quality, in terms of FZDs, is difficult to measure directly during the welding process unless production is stopped which is nearly infeasible as it is economically unjustified; whereas, it is measured by signals and relating these signals to relevant KPIs such as FZDs.
Therefore, development of in-process monitoring methodologies is essential to assure the weld quality. The quality assurance consists of two stages: (i) process monitoring, and(ii)process control. The process monitoring is the manipulation of measurements in determining the current state of the welding processes; whereas, the process control is the manipulation of process parameters based on the information gathered from the monitoring stage in order to regulate the processes.
Figure 1.5 shows the outline of the dissertation consists of two major com- ponents as follows:
(a) The process parameters are defined as the parameters required to setup a welding process and they can be divided into two categories: (i) controllable parameters are those that can be varied during welding (e.g. laser power, weld-
ing speed, focal offset,etc.), and(ii) uncontrollable parameters are those that cannot be modified during welding, such as joining gap. In this dissertation, the laser dimpling process (upstream process) is utilized to develop a joining gap between two sheet metals. A methodology is developed inChapter 4to select robust the laser dimpling process parameters in the presence of process variation.
(b) The primary aims of the proposed framework are to consistently produce ade- quate quality welds by monitoring FZDs and to control the process parameters that affect FZDs. The effectiveness of the proposed framework is dependent on the prediction of the process model. It is the mathematical representa- tion of the actual process (i.e. remote laser welding process) which could be theoretical, empirical or simulation based.
Theoretical models are based on analytical solutions of governing physical equations, empirical models are developed from experimental design methods for example response surface methodology, and simulation models are numeric solutions of the governing physical equations with the help of computers. It should be noted that the computational time and accuracy of the developed model should be as close as possible to the process time and the output (e.g. macro-section image of the weld) to utilize the developed model in the pro- posed framework. A physics-driven model based on the occurring physical phenomena is developed inChapter 5by considering the computational time and the accuracy of the prediction.
Process Control RLW Process
KPI1:Penetration (PT) KPI2:Top Width (TW) KPI3:Interface Width (IW)
FZ Profile KPI Requirements Process Adjustment Towards Zero Weld Defects KPI Extraction Welding Model Correction Model Fixturing Design (Maximum Joining Gap Control) Process Parameters Beam Parameters Laser Dimpling Process (Chapter 4) (Minimum Joining Gap Control) Process Emissions (a) (b)
Fig. 1.5. The outline of the dissertation: (a) controlling minimum joining gap require- ment in the RLW process by utilizing the laser dimpling process (Chapter 4), (b) the physics-driven process model for quality assurance for the RLW process (Chapter 5)
Based on the aforementioned research scopes, the definition of the research questions (RQ) are outlined as:
RQ 1 How to select robust dimple process parameters to achieve given quality re- quirements in the presence of process variation?
The scope of this thesis is the laser welding of galvanized sheet metal which is highly utilized in the BIW due to its corrosion resistance, strength, cost and hardness (Ma et al., 2012; Zhao et al., 2012; Chen et al., 2013). However, the laser welding of this metal is unstable and difficult to control because of the vaporization temperature of the zinc (∼907◦C) is lower than the melting tem- perature of the steel (∼1500◦C) resulting in highly pressurized zinc vapour on the faying surfaces during the welding process. Left unaddressed, such zinc vapour can easily be trapped inside the molten pool which can lead to welding defects such as porosity, spatter, burn-through, and severe undercuts (Norman et al., 2009; Chen et al., 2013).
Over the past few years, various methodologies have been developed to miti- gate zinc vapour from the welding medium without causing any disturbance in the molten pool and the keyhole. The state-of-the-art method in the auto- motive industry is the ventilation of zinc vapour through a joining gap. The required gap can be developed by, for example, “laser dimpling process”. It is a very promising manufacturing process as it does not require any additional equipment and suppliers. Additionally, the same laser source and the fixture adopted for welding can be utilized. In this process, dimples with a height of 0.05 mm - 0.2 mm are produced on the surfaces by the rapid movement of the laser beam at a short distance. The number of dimples and the position depends on the weld seam (i.e. linear or circular weld). After the realization of dimples, two sheet metals are placed in overlap configuration and welding is performed.
Dimples work as a spacer between two sheet metals which control the mini- mum joining gap. On the other hand, the maximum joining gap is controlled by the welding fixture (Das et al., 2015). Since the joining gap is an un- controllable process parameter and achieved by an upstream process, a novel methodology is introduced to control dimples so that minimum joining gap is always achieved in the presence of process variation to prevent weld defects, such as blow-hole, spatters,etc.
According to the reviewed literature in Section 3.1, the research gap is identi- fied as: (i) lack of KPIs to determine the dimple quality, (ii) complete char- acterisation of the dimpling process since the existing literature have focussed mainly on single-input (i.e. welding speed), single-output (i.e. dimple height) scenario (SISO scenario), and (iii) selection of process parameters for given quality requirement in the presence of the process variation.
RQ 2 How to directly monitor multi Fusion Zone Dimensions (Penetration, Top Width, Interface Width) in the overlap laser welding of galvanized steels to assess the weld quality?
Currently, data-driven process models are widely utilized for in-process mon- itoring. These models work according to the principle of the acquisition of data (i.e. acoustic, optical and visual emissions) using sensors then correlated them using multivariate statistical methods and machine learning algorithms to the formation of weld defects. The most common sensors in use today for in-process monitoring are photodiode (Eriksson et al., 2010), high-speed and thermal cameras (Kawahito et al., 2009; Tenner et al., 2015).
According to the reviewed literature about process monitoring, which is given in Section 3.2, the limitations of current data-driven in-process monitoring methods are that sensor signals are multi-dimensional and multi-modal, it is often not realistic to use them directly as an input for control algorithms. They do not monitor directly weld defects, but instead they monitor signals arising from the process and develop predictive models. As a result, changes in process parameters or material properties can be handled only by rebuilding these predictive models. Moreover, they are capable of detecting external KPIs but insufficient to directly monitoring internal KPIs, especially FZDs such as interface width.
An alternative to the data-driven process model is a physics-driven model which numerically solves the governing physics in the laser welding process. The developed model emulates the transverse cross-section of the weld. How- ever, the problem is to obtain simulation results at a given accuracy within the welding process time. Therefore, a simplified model is developed that al- lows a fast estimation of FZDs namely; penetration, top width; and interface width. The key idea is to integrate the physics-driven model with gathered data to reduce computational time without losing accuracy. Furthermore, the physics-driven model consists of four steps as: (i) calculating laser intensity acting on the material, (ii) calculating keyhole profile in using an analytic method, (iii) solving the heat equation using FEM to calculate fusion zone (FZ) profile, and (iv) aforementioned FZDs are obtained from the calculated FZ profile.