8 Industry 4.0 in Welding
269
8.3.2 Indirect Monitoring
The indirect monitoring utilizes a state variable as a measure of the quality variable;
i.e. it refers to the sensing of various physical quantities; the values of which are
analysed to indirectly correlate with the weld quality. The necessity of the indirect
approach of monitoring aims to address the difficulties related to direct monitoring
techniques. These physical quantities are: current, power, force, torque, temperature,
acoustic emission, sound, vibration, etc. Since attempts are made to correlate the
weld quality with a physical quantity, the result is less accurate as compared to
that of the direct monitoring approach. However, the indirect monitoring technique
is more practical, economic and suitable for industrial applications. The biggest
advantage of the indirect monitoring techniques is the in-line engagement with the
manufacturing process. This provides the real-time capability. In order to build an
efficient indirect monitoring system, it is crucial to identify the quality variable,
and the suitable signal to identify/predict the quality. Accordingly, a sensor can be
selected. The acquired signal is then processed online or offline using various digital
signal processing techniques. In order to predict the quality, ML models are utilized.
This prediction of the weld quality forms a decision. This decision is often compared
with a reference value, to determine the deviation or error, and accordingly a suitable
control system is designed for sending feedback to the process. The utility of each
of these signals in welding has been discussed below.
8.3.2.1 Current/voltage and Power Signals
In arc welding, fusion occurs because of the heat derived from the conversion of
electrical power. The arc is established between two conductors and sensing this arc
helps in deriving useful information about the welding technique. The sensing of
arc refers to capturing of the signature of the electrical parameters, such as current
or voltage [41]. In case of robotic arc welding, the electrical parameters have been
utilized for seam tracking [33]. In GMAW, the current signal has been found to be
one of the most low-cost solutions for tracking the weld seam. This is because of the
strong negative correlation existing between the arc current and arc length. Further,
the lateral location of the weld seam has also been determined with the help of current
signature [33]. The current signature has been found to have a peak whenever the
torch is at the edge and valley while the torch passes over the seam. The voltage
signature has also been utilized for seam tracking [33]. Studies have utilized these
electrical parameters for predicting the weld quality. For instance, the joint strength
of the weld has been predicted by using features extracted from the current and
voltage signals [42, 43]. These studies also report the utility of studying the signal
in the time domain [42], and also in the time–frequency domain [43]. The root mean
square (RMS) value of current and voltage signals in the time domain, along with the
welding parameters, was fed to a ML model for prediction of the weld strength. In
the time–frequency domain, the RMS values of the coefficients of different wavelet
269
8.3.2 Indirect Monitoring
The indirect monitoring utilizes a state variable as a measure of the quality variable;
i.e. it refers to the sensing of various physical quantities; the values of which are
analysed to indirectly correlate with the weld quality. The necessity of the indirect
approach of monitoring aims to address the difficulties related to direct monitoring
techniques. These physical quantities are: current, power, force, torque, temperature,
acoustic emission, sound, vibration, etc. Since attempts are made to correlate the
weld quality with a physical quantity, the result is less accurate as compared to
that of the direct monitoring approach. However, the indirect monitoring technique
is more practical, economic and suitable for industrial applications. The biggest
advantage of the indirect monitoring techniques is the in-line engagement with the
manufacturing process. This provides the real-time capability. In order to build an
efficient indirect monitoring system, it is crucial to identify the quality variable,
and the suitable signal to identify/predict the quality. Accordingly, a sensor can be
selected. The acquired signal is then processed online or offline using various digital
signal processing techniques. In order to predict the quality, ML models are utilized.
This prediction of the weld quality forms a decision. This decision is often compared
with a reference value, to determine the deviation or error, and accordingly a suitable
control system is designed for sending feedback to the process. The utility of each
of these signals in welding has been discussed below.
8.3.2.1 Current/voltage and Power Signals
In arc welding, fusion occurs because of the heat derived from the conversion of
electrical power. The arc is established between two conductors and sensing this arc
helps in deriving useful information about the welding technique. The sensing of
arc refers to capturing of the signature of the electrical parameters, such as current
or voltage [41]. In case of robotic arc welding, the electrical parameters have been
utilized for seam tracking [33]. In GMAW, the current signal has been found to be
one of the most low-cost solutions for tracking the weld seam. This is because of the
strong negative correlation existing between the arc current and arc length. Further,
the lateral location of the weld seam has also been determined with the help of current
signature [33]. The current signature has been found to have a peak whenever the
torch is at the edge and valley while the torch passes over the seam. The voltage
signature has also been utilized for seam tracking [33]. Studies have utilized these
electrical parameters for predicting the weld quality. For instance, the joint strength
of the weld has been predicted by using features extracted from the current and
voltage signals [42, 43]. These studies also report the utility of studying the signal
in the time domain [42], and also in the time–frequency domain [43]. The root mean
square (RMS) value of current and voltage signals in the time domain, along with the
welding parameters, was fed to a ML model for prediction of the weld strength. In
the time–frequency domain, the RMS values of the coefficients of different wavelet