^{*}

This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).

The present study shows the relationship between welding quality and optical-acoustic emissions from electric arcs, during welding runs, in the GMAW-S process. Bead on plate welding tests was carried out with pre-set parameters chosen from manufacturing standards. During the welding runs interferences were induced on the welding path using paint, grease or gas faults. In each welding run arc voltage, welding current, infrared and acoustic emission values were acquired and parameters such as arc power, acoustic peaks rate and infrared radiation rate computed. Data fusion algorithms were developed by assessing known welding quality parameters from arc emissions. These algorithms have showed better responses when they are based on more than just one sensor. Finally, it was concluded that there is a close relation between arc emissions and quality in welding and it can be measured from arc emissions sensing and data fusion algorithms.

Gas metal arc welding—GMAW in short circuit transfer mode (GMAW-S), is a manufacturing process widely used in the metallic construction industry. Diverse advantages such as the high rate of metal transference, elevated penetration and facility for welding in diverse positions, makes this process the most widely used. One way of defining welding quality is through standard specifications which list the limits of discontinuities which are acceptable for a particular application. Quality specifications are not the same for all weld applications. A weld acceptable for static loading may not be acceptable for a dynamic loading application. Six items for assuring the weld quality must be considered: process selection, joint penetration, procedure, pretesting, qualified personnel and in-process monitoring [

Many efforts have been encouraged by the industry in order to guarantee welding quality. One of them is the on-line monitoring of some welding parameters which reduces the severity and time requirements of the quality control tests. Classically, the arc tension and welding current are monitored. These parameters are electric arc stability indicators and their behavior also has direct implications in the heat and metal transference which is reflected in the weld bead geometry. High stability in welding does not necessarily mean high quality. Welding quality, in addition to stability, involves other requirements according its application, but certainly the stability is an essential condition. Beside voltage and current parameters, during arc welding operations, the electric arc produces electromagnetic and mechanical emissions observed as magnetic fields, luminescence, light flashing and sound (named as arc emissions). Typically welders use these emissions in combination with their knowledge as feedback information for controlling the welding process aiming to achieve high quality. Different researches shows that is possible to detect some interferences and assess the welding quality by measuring acoustic and optical arc emissions [

The electric arc is a current flowing between two electrodes through an ionized column of gas called a plasma. The space between the two electrodes can be divided into three areas of heat generation: the anode, the cathode and the arc plasma [

Infrared emission is originated by the electromagnetic energy emitted by the welding arc and sensed just at the infrared wavelength (0.8–1.1 μm specified in the pyrometer datasheet). Its intensity and wavelength of energy produced depends on the welding parameters, electrode and base metal composition, as well as the fluxes of shielding gas. The intensity of this electromagnetic emission I_e is governed by Planck's law which describes the spectral radiance of unpolarized electromagnetic radiation at all wavelengths emitted from a black body at absolute temperature T. As a function of frequency v, Planck's law is written as:

In _{e} is also named as spectral radiance (jm^{2}sr^{−1}), T temperature (k), v frequency (HZ), h Plank constant (6.62606896 × 10^{−34} Js), c speed of light (3.0 × 10^{8}m/s) and k Boltzmann constant (≈1.3806504 × 10^{−23})J/k).

In the GMAW-S process, the metal is transferred to the welding pool when the molten tip of the consumable electrode contacts the molten puddle. This generates sudden changes in the power of the welding arc. In GMAW-S, the welding arc is characterized by ignitions and extinction sequences and the welding arc sound fits this welding arc behavior. In each arc ignition there is a sound peak as well as when the arc has been extinct, a small sound peak is produced (see _{e}(t) and the welding arc power P(t)=V(t) I(t)could be expressed by

where K is a proportionality factor, α is a geometrical factor, γ the adiabatic expansion coefficient of air and c the velocity of sound in the arc (

Stationarity is a statistical property of random nature signals which means that the statistical quantities are independent of the absolute time and dependant only on relative times, in other words a signal is stationarity when its essential statistical properties are invariant over time. Two kinds of stationarity are distinguished: weak and strong stationarity. Weak stationarity is meant when the first and second moments are independent of time and constants, that is, 〈E_{t}〉 = μ and 〈| E_{t} − μ |^{2}〉= σ ^{2}, (where 〈 〉 stands for the ensemble average). For finite random signals that is the case of the welding arc emissions, the behavior of the mean value and variance cannot be enough estimators for stationarity. A stochastic process {E_{t}} with t as an integer number, is denominated as strongly stationary if any set of times t_{1},t_{2} and any integer k the joint probability distributions of

Probability average:

Time Average:

Fluctuations:

Since

The time average of the square of the fluctuations is evaluated by using

Finally the autocorrelation is defined as:

It is more convenient to work with the normalized autocorrelation function AcF_{E′} defined in _{E′} =1 indicates weak stationarity and AcF_{E′} =0 indicates strong stationarity

Note that c_{E′} = 1 indicates weak stationarity and c_{E′} = 0 indicates strong stationarity.

Generally, the autocorrelation is expected to decay exponentially, and the fluctuations are expected to become uncorrelated after a sufficiently long-time. In the above figures it is observed that autocorrelation functions tend to zero, which means that both welding arc emissions have a strong stationarity after a certain time and therefore they can be used as welding monitoring parameters.

The welds were carried out on steel plates AISI 1020 (140 mm × 101.2 × 9.60 mm) using AWS A5.18 ER70S-6 1.2 mm in diameter electrode wire; the shield gas was the mixture of argon and carbonic anhydride M21 (ATAL 5A/Ar 82% + CO_{2} 18%). The welding runs were performed maintain a fixed contact tip work distance—CTWD at 10 mm and shield gas flow at 15 L/min. These experiments were executed setting combinations for four arc voltage levels (18, 19, 20 and 21 V), five levels to wire feed speed (3.0, 3.5, 4.0, 4.5 and 5.0 m/min) and three welding speed levels (7, 9 and 11 mm/s) which in total gives sixty welding experiments.

In data fusion theory, there are three principal architecture topologies that are categorized according to the type of sensor configurations: complementary, competitive and cooperative. In this work, the competitive topology was used (see

These set equations are based on third deviation rule; the expression

In

Before the application of data fusion concept, it is necessary modeling the quality level signals. Each signal could be considerate as time series and it is modeled as a parametric model:
_{t+1} is the transition matrix from the state X_{t} to X_{t+1}, X in this work, is the quality level parameter and W is the noise, represented as a random variable with normal distribution with zero mean and variance Q.

The model for the quality level measurer is shown in

The signal given by the sensor (m_{t}) is the quality level measured by each arc emission sensor (X_{t}) with an added noise (V_{t} = V_{ex} + V_{in}) The noise also is a random variable with normal distribution with zero mean, but variance Q. With this model it is performed an overall quality assessment system by data fusion.

Data fusion is the process of combining and integrating measured features originated from different sensors to produce more specific, comprehensive, and unified information about a monitored process such as the arc welding features in the case of this paper. Diverse parameters are involved in many production processes. Measuring their behavior is very important for achieving a high quality production. Some measured parameters m_{t} are used by workers for visualization (or settings) of the production line and other parameters are used as feedback variables for process control systems. The measurement systems are composed of transducers and sensors which measure and read diverse variables and parameters of production processes. Along with signals of parameters, X_{t} undesirable noise signals V_{ex} are also measured (external noise) and both signals are altered by circuits of conditioning, transmission and calibration; These signal management circuits also adds V_{in} noise on the measured signals (internal noise). The external noise V_{ex} has intrinsic nature in the processes as well as the internal noise V_{in} in the measurement systems and both signals V_{ex}+V_{in} are responsible for the errors of measurement. Finally, the measured signal m_{t} is constituted of three components: the measured variable, external noise and internal noise (m_{t}= X_{t} + V_{ex} + V_{in}).

Data fusion in sensing (named also as multi-sensor fusion) are a set of techniques broadly used in science areas such as image processing, remote sensing, sensor networks,

Multi-sensor measurement systems offer numerous advantages over single sensors when it comes to the fundamental tasks of utilizing and delivering information for a specific objective [

State and covariance time propagation:

State and covariance measurement update:

First, the state estimate is generated by processing the measurement data from each sensor. Fusion is obtained by combining the state estimates using a weighted sum of the two independent state estimates. The weight factors used are the appropriate covariance matrices. Thus, these state estimates and the corresponding covariance matrices are fused as follows, that is, the fused state and covariance matrix are computed using the following expressions:

Here,

A quality assessment system based on monitoring of arc welding emissions and data fusion was performed. The data fusion process has shown positive results detecting induced perturbations throughout the welding path in comparison at usual quality assessment methods based on single sensoring. Many researchers obtain quality level models as a time series or mechanistic models becoming the quality assessment system dependent on some constants that are usually obtained experimentally, which makes the assessment system unreliable. This limitation is related to the lack of relationships between welding quality models and the welding parameters and this drawback is avoided by using multiple sensoring techniques. By using data fusion of quality levels, the capability and sensitive of the overall quality assessment system were improved.

By monitoring arc welding emissions it was possible to detect induced perturbations during the welding runs. Some perturbations are detected by the acoustic emissions and others by infrared emissions. Acoustic monitoring was sensitive to environmental noise and the quality level extracted from it, has higher ripples than the quality level sensed through infrared emissions. Sensoring based on data fusion improves the monitoring of the welding quality and it could be an alternative to classical on-line methods of assessment and inspection used for detecting and finding disturbances that are based in direct measurements of parameters such as arc voltage, welding current, wire feed speed, and others.

The authors gratefully acknowledge support for this Project from The Brasilia University and the CNPq (Brazilian Research Council).

Welding arc parameters and emissions.

Welding arc emissions autocorrelation.

Experimental Setup.

Competitive topology.

Distribution of RMS and Short circuit rate signal.

Pre-processing data signal stages, (

Quality level parameters, (

Detailed data fusion architecture.

Quality level parameters, (