8 Industry 4.0 in Welding
281
research shows the application of multi-sensor fusion in welding processes [54, 63,
77, 106]. A challenge in the sensor fusion task is to deal with the heterogeneous data.
The detection of different types of phenomena and producing measurements for
several attributes makes it cumbersome to fuse the information gathered. In addition,
the sampling rate is not same for different sensors which make it difficult for real-time
application. Even if the data are sampled at the same frequency, a delay is introduced
in the process, owing to the volume of data being generated from these sensors.
There are several signal processing techniques which have been discussed in the
preceding sections. The selection of a proper technique for extraction of meaningful
information is vital [52]. An article shows the importance of multiple sensors such as
optical, acoustic signal and arc voltage, for monitoring of GTAW technique [107]. A
feature-level data fusion was applied, and a SVM model was built for weld classification. Likewise, in case of the spot welding technique, features from multiple signals,
namely temperature, current, voltage, ultrasonic, force and dynamic resistance, were
extracted and fed to a ML model for predicting the weld quality [77]. Another study
demonstrated the importance of using multiple sensors for process monitoring over
a single sensor, in FSW technique [54]. The sensors utilized in this study were force,
torque, and power, which were fused in a ML model for predicting the weld quality.
The accuracy of prediction was high in case of multiple sensors as compared to the
information gathered from a single sensor.
AI and ML have been contributing significantly towards the automation of production processes. The most promising factors for the implementation of AI are the transformation of human knowledge and skill into computer software. AI will provide
engineering assistance for the automation of low-level manufacturing issues. Over
the years, the integration of machine tools, monitoring algorithms and machines
with computer network have led to enhanced product quality. The control mechanisms that govern the manufacturing processes in every step are: sensory systems,
data mining, ML models, AI and expert systems. One of the essential requirements in
manufacturing is the decision on the most suitable or the optimal combination of evaluation criterion. Sufficient information does not exist to determine the best possible
combination of design parameters. In case of non-automated system, the experienced personnel in the organization is the most sought after during decision-making
process. AI involves expert systems, which once fed with adequate information are
capable of taking care of decisions at the shop floor. The intelligent and automated
machines working on AI can be utilized by integrating these machines with the intelligent controlling systems that can monitor and diagnose the ongoing operations.
This will lead to better product quality along with competitive advantage. The ML
models have been utilized for classification of welds, pattern recognition, prediction
of weld quality and health monitoring. A review on the use of various ML models
in monitoring of the arc welding process can be found in Ref. [108]. This review
suggests the artificial neural network (ANN) to be the best suited for dealing with the
noisy and nonlinear data. A recent article reviewed the application of ML models in
monitoring of the laser welding technique [109]. This review covered models such
as feedforward NN, backpropagation NN, and SVM. Apart from these three, there
281
research shows the application of multi-sensor fusion in welding processes [54, 63,
77, 106]. A challenge in the sensor fusion task is to deal with the heterogeneous data.
The detection of different types of phenomena and producing measurements for
several attributes makes it cumbersome to fuse the information gathered. In addition,
the sampling rate is not same for different sensors which make it difficult for real-time
application. Even if the data are sampled at the same frequency, a delay is introduced
in the process, owing to the volume of data being generated from these sensors.
There are several signal processing techniques which have been discussed in the
preceding sections. The selection of a proper technique for extraction of meaningful
information is vital [52]. An article shows the importance of multiple sensors such as
optical, acoustic signal and arc voltage, for monitoring of GTAW technique [107]. A
feature-level data fusion was applied, and a SVM model was built for weld classification. Likewise, in case of the spot welding technique, features from multiple signals,
namely temperature, current, voltage, ultrasonic, force and dynamic resistance, were
extracted and fed to a ML model for predicting the weld quality [77]. Another study
demonstrated the importance of using multiple sensors for process monitoring over
a single sensor, in FSW technique [54]. The sensors utilized in this study were force,
torque, and power, which were fused in a ML model for predicting the weld quality.
The accuracy of prediction was high in case of multiple sensors as compared to the
information gathered from a single sensor.
AI and ML have been contributing significantly towards the automation of production processes. The most promising factors for the implementation of AI are the transformation of human knowledge and skill into computer software. AI will provide
engineering assistance for the automation of low-level manufacturing issues. Over
the years, the integration of machine tools, monitoring algorithms and machines
with computer network have led to enhanced product quality. The control mechanisms that govern the manufacturing processes in every step are: sensory systems,
data mining, ML models, AI and expert systems. One of the essential requirements in
manufacturing is the decision on the most suitable or the optimal combination of evaluation criterion. Sufficient information does not exist to determine the best possible
combination of design parameters. In case of non-automated system, the experienced personnel in the organization is the most sought after during decision-making
process. AI involves expert systems, which once fed with adequate information are
capable of taking care of decisions at the shop floor. The intelligent and automated
machines working on AI can be utilized by integrating these machines with the intelligent controlling systems that can monitor and diagnose the ongoing operations.
This will lead to better product quality along with competitive advantage. The ML
models have been utilized for classification of welds, pattern recognition, prediction
of weld quality and health monitoring. A review on the use of various ML models
in monitoring of the arc welding process can be found in Ref. [108]. This review
suggests the artificial neural network (ANN) to be the best suited for dealing with the
noisy and nonlinear data. A recent article reviewed the application of ML models in
monitoring of the laser welding technique [109]. This review covered models such
as feedforward NN, backpropagation NN, and SVM. Apart from these three, there