The Advanced Real-Time Monitoring of New Welding Processes in the Aircraft Industry ^†

Abstract

While the monitoring techniques employed are well established in other fields, their application to the novel processes of ultrasonic and resistance welding of composites is innovative. This study details the adaptation of these techniques to monitor the essential parameters that influence the quality of welds in composite materials within the context of the MFFD (Multi-Functional Fuselage Demonstrator) during the WELDER project. This article presents findings from the application of a sophisticated monitoring system to new welding processes for composites, demonstrating significant enhancements in process control, quality assurance, and operational safety. In ultrasonic welding, temperature, speed, power, amplitude, and displacement were monitored, whereas in resistance welding, voltage, load, and temperature were tracked. This comprehensive data collection is crucial due to the sensitivity of composite materials to welding parameters, which directly affect the integrity and performance of the final product. The data transmission utilized the OPC UA protocol, ensuring secure, reliable real-time data flow to visualization interfaces in safe environments. The integration of real-time monitoring and advanced data analysis helps in identifying potential defects during the welding process, thereby enhancing both the reliability and efficiency of composite welding. The results underscore the potential of these technologies to advance the manufacturing practices for high-performance composite materials in the aeronautics field.

1. Introduction

The integration of carbon fiber-reinforced thermoplastic composites (CFRTPs) in airframe manufacturing is transforming standards in the aerospace industry. Chosen for their ability to be efficiently welded and heat reformed, CFRTPs offer significant advantages in terms of manufacturability and maintainability. Compared to traditional metals, they provide superior strength and remarkable weight reduction, which facilitates more innovative designs and improves sustainability due to their greater recyclability.

Within this context of innovation, the WELDER project, developed under the Multi-Functional Fuselage Demonstrator (MFFD) framework, plays an important role in the improvement of ultrasonic and resistance welding. This study focuses on how specialized monitoring systems, implemented in weld heads, are critical to optimize the quality and efficiency of the welding process.

The architecture of the monitoring software and hardware, which uses the OPC UA protocol to ensure secure and efficient communication of data in real time, is an integral part of these systems. These advances not only optimize airframe manufacturing, but by integrating artificial intelligence to calculate bond strength, they also usher in a new phase of innovation in aircraft production. This approach has a direct impact on cost efficiency and improved operational sustainability.

2. State of the Art

2.1. Ultrasonic Welding and the Application of Machine Learning

The ultrasonic welding of thermoplastic composites has been the subject of studies that seek to optimize process parameters using advanced technologies. Important research in this area explores new input factors for machine learning that can predict ultrasonic weld quality, which is vital for the implementation of more accurate process controls [1]. The use of thermography and acoustic emission as input factors for machine learning algorithms to predict the static ultrasonic welding quality of thermoplastic composite materials was explored by [2], showing promising results in the classification of joint quality. Other relevant studies have analyzed the impact of sonotrode alignment on joint strength and reliability, emphasizing the need for accurate positioning to prevent defects and ensure consistent weld quality [3]. Additionally, robust process control strategies have been proposed to ensure consistent weld quality in the sequential ultrasonic spot welding of thermoplastic composites. Research has demonstrated the importance of controlling key parameters, such as energy input and process monitoring, to maintain uniform joint strength, which is critical for high-performance composite applications [4]. Furthermore, continuous ultrasonic welding has been investigated to improve weld uniformity. Studies have shown that replacing a thin film energy director with a woven polymer mesh significantly enhances the weld quality by promoting better contact and ensuring uniform heat distribution throughout the weld seam [5].

2.2. Resistance Welding and Process Optimization

Resistance welding has also experienced significant advancements in its application to thermoplastic composites, particularly in the aerospace industry. Research has demonstrated the feasibility of using resistance welding for complex aeronautical structures, focusing on optimizing parameters, such as current, pressure, and time [6]. Additionally, advanced statistical techniques, including response surface methodology and regression models, have been employed to refine process variables, achieving enhanced weld performance and reducing variability [7]. Moreover, the influence of temperature on the strength of resistance-welded joints has been studied, showing that the lap shear strength decreases with increasing temperature, except for a stable region between 50 °C and 90 °C. The main failure mechanisms were identified as glass fiber/matrix de-bonding, highlighting the importance of the fiber/matrix interfacial strength, and stress distribution at the joint overlap [8].

2.3. Advanced Monitoring in Composite Welding

The integration of advanced monitoring systems has revolutionized composite welding processes, with technologies such as OPC UA playing a critical role in ensuring real-time data acquisition and process control. OPC UA’s secure and interoperable architecture enables seamless communication between devices, facilitating the monitoring of parameters like temperature, load, and displacement. Studies have demonstrated its effectiveness in implementing 3D monitoring systems to enhance process supervision and defect detection [9,10]. Furthermore, integrating OPC UA with Time-Sensitive Networking (TSN) has shown significant improvements in synchronizing real-time data across heterogeneous manufacturing systems [11].

2.4. Review of Machine Learning Methods in Manufacturing Processes

Machine learning has emerged as a transformative tool in the optimization of manufacturing processes, enabling the analysis of large datasets to uncover patterns and insights. A systematic review [12] highlights its applicability across various domains, including composite welding, where it is used to predict process outcomes, detect anomalies, and optimize parameter selection. Recent studies have demonstrated its use in robotic welding, where advanced algorithms, such as support vector machines combined with particle swarm optimization (SVM-PSO), have been applied to model and refine welding parameters, achieving significant improvements in weld quality and process efficiency [13]. Similarly, convolutional neural networks (CNNs) have been employed for real-time quality prediction in gas metal arc welding (GMAW), demonstrating their ability to assess weld quality based on process images and sensor data [14]. More broadly, artificial intelligence (AI), including machine learning, is playing an increasingly pivotal role in advanced manufacturing. As noted in a recent study [15], AI applications extend beyond welding to encompass areas such as product design, process planning, and predictive maintenance, offering enhanced efficiency, adaptability, and precision. However, challenges such as data quality, infrastructure requirements, and organizational resistance remain barriers to widespread adoption, underscoring the need for industry–academic collaboration to realize its full potential.

3. Methodology

The WELDER project, integrated in the Multi-Functional Fuselage Demonstrator (MFFD), focuses on advancing the welding of thermoplastic composites through an integrated approach that includes an innovative weld head design and the optimization of welding parameters. This approach is complemented by an advanced monitoring system to ensure the quality and integrity of welded joints. Initially, extensive laboratory tests were conducted using various sensors and monitoring techniques tailored to ultrasonic and resistance welding processes. The data collected from these tests were meticulously analyzed to understand how different parameters affect weld quality, thus guiding subsequent weld head design.

3.1. Design and Optimization of Welding Heads

3.1.1. Ultrasonic Welding

Based on the data analysis, a thermal camera was positioned to monitor not only the material temperature but also to verify the sonotrode alignment. This requires the camera to observe the temperature before and after the cooling block (Figure 1). At the same time, the signals from the generator and the positionings of the different axes are read from the PLC.

3.1.2. Resistance Welding

A PCB was designed to integrate temperature voltage and load measurements. This PCB can read the temperature of 20 thermocouples placed in the head, the value of the load cells located in the head, and, finally, it has a voltmeter to measure the voltage.

3.2. Data Management and Real-Time Monitoring

All collected data are initially stored in CSV files to facilitate later review and analysis by technicians. In addition, the data, including images and operating values of the PLCs controlling the welding heads, are published on an OPC UA server. This allows real-time access through monitoring interfaces, facilitating continuous supervision and the ability to make immediate adjustments in response to any indication of problems or inefficiencies. In addition, this OPC UA server allows multiple connections in a secure manner, which may include digital twins, visualization interfaces, or simulation systems. (Figure 2).

3.3. Detailed Data Analysis

• Speed and Y-Axis Monitoring in Ultrasonic Welding: Motor torque and Y-axis displacement are monitored to detect any unwanted contact between the sonotrode and the material, which could cause damage. Y-axis information and a decrease in speed and an increase in displacement indicate unwanted contact. It should be noted that the Y-axis measurement not on the large-scale demonstrator does not return usable values when binding occurs in the structure.
• Thermocamera Data Analysis: The thermocamera is used to monitor whether the material reaches or exceeds the melting temperature. If, during the cooling block measurement, it is observed that the temperature is higher than the previous one at the same point, it indicates that the melting temperature has been reached internally. In addition, it allows us to verify if the sonotrode is correctly aligned during the process (Figure 3).
• Temperature and Load Data in Resistance Welding: The thermocouples, although they are not in direct contact with the welding head and show delayed values, verify whether the energy conductor is well positioned during the process. For this purpose, the differences between the temperature of the thermocouples are analyzed. Poor positioning does not mean poor welding. The load data help to detect the precise moment when the melting of the material begins, which is important to know the time needed for the proper administration of the applied energy (Figure 4).
• Resistance Welding Voltage Data Analysis: Voltage monitoring is critical to ensure that the applied energy is adequate. The data are analyzed to ensure that the voltage is neither too low, which would prevent welding, nor too high, which could cause damage to the power conductor (Figure 5).

3.4. Use of Artificial Intelligence

For both welding techniques, artificial intelligence models were developed and trained with the objective of optimizing the process parameters and predicting the strength of the welded joints. The common process and specific differences in feature extraction for each type of weld are detailed below:

3.4.1. Feature Extraction

• Ultrasonic Welding: temperature and power of sensory data were mainly used.
• Process Extraction: extracted features included direct values from temperature and power measurements, leveraging these data to capture the essential dynamics of the welding process.
• Resistance Welding: Focused on polynomial fitting to perform resizing of signals, and extraction of features such as maxima, minima, and slopes during the heating process. Polynomials were fitted to the signals to model the behavior during heating and basic statistical characteristics that reflect critical properties of the process, such as trend and variability of the signals, were calculated.

3.4.2. Training and Validation

• Models Used: convolutional neural networks were implemented for both processes due to their effectiveness in handling complex spatial and temporal characteristics of the weld data.
• Data Splitting: data were split 70% for training and 30% for validation, ensuring a rigorous and realistic evaluation of the models on unseen data.
• Evaluation: Standard metrics, such as R², MAE (Mean Absolute Error), and MSE, were used to evaluate the accuracy and reliability of model predictions. This provided a quantitative basis for comparing the effectiveness of the model in predicting weld quality.
• Visualization of Results: the model predictions and actual values for both welding methods were displayed comparatively, providing a direct visual assessment of the predictive capability of the models and their applicability in real production environments.

4. Results

The analytical component of the project focused on optimizing welding parameters and accurately detecting conditions that could lead to failures during the welding process. Specifically, artificial intelligence was applied to predict the bond breaking strength of the welds. Machine learning models were trained using data from previous welds, applying advanced regression and classification techniques. They were evaluated using metrics such as mean absolute error (MAE), mean square error (MSE), and the coefficient of determination (R²).

4.1. Artificial Intelligence Results in Ultrasonic Welding

As illustrated in the graph provided (Figure 6), the model has demonstrated a high ability to predict weld resistance, expressed in megapascals. The blue solid line represents the actual resistance values, while the orange dashed line shows the values predicted by the model. The results highlight a coefficient of determination (R²) of 0.9360, indicating a high degree of accuracy in the model predictions. The accuracy of the predictions and the low mean square error (MSE of 1.72) and mean absolute error (MAE of 1.04) reinforce the reliability of the model in practical applications.

4.2. Artificial Intelligence Results in Resistance Welding

The AI models applied in resistance welding demonstrated exceptional accuracy, with an R² Score of 0.96 and a Mean Absolute Error (MAE) of 0.24. These results indicate that the models are extremely effective in predicting welding results, ensuring the quality and efficiency of the process. The near-perfect alignment between the actual and predicted values, as visualized in Figure 7, underscores the model’s ability to operate effectively under real production conditions.

The use of real-time monitoring has enabled precise adjustments to the welding parameters, resulting in a significant reduction in the defect rate and improving operational efficiency. The ability to intervene quickly for any deviation from established parameters has been important in minimizing material waste and ensuring that each joint meets the highest quality standards. For example, during ultrasonic welding processes, it is possible to observe when the sonotrode becomes misaligned and stops the process. In resistance welding, you can observe when the energy director starts to melt and monitor the time based on this point.

5. Conclusions

The implementation of real-time monitoring and data analysis technologies in the WELDER project has been shown to improve the efficiency and quality of thermoplastic composite welds:

• Cost Reduction: the application of data analytics to predict bond break strength can result in a marked reduction in NDT testing.
• Operational Efficiency: The ability to monitor and adjust processes in real time has been invaluable. Operational efficiency has been especially improved by predicting process failures, which has allowed intervention before the material was compromised, thus avoiding wasted material. This system has minimized interruptions in the production process, improving operational efficiency and reducing associated costs.
• Innovation in Manufacturing Techniques: The WELDER project has demonstrated how the integration of monitoring and data analysis technologies into traditional manufacturing processes can facilitate innovation. The techniques developed and refined in this project have the potential to be applied in other industrial sectors, encouraging a greater adoption of advanced manufacturing solutions.
• The results obtained underscore the need to continue exploring the integration of digital technologies in manufacturing. Future research should focus on the adaptability and scalability of these monitoring systems for different materials and production environments, with the aim of further improving the accuracy and efficiency of automated processes.

In conclusion, the WELDER project has provided valuable insights into the application of advanced analysis and monitoring techniques in composite welding, demonstrating that the fusion of state-of-the-art technology with traditional manufacturing techniques is not only possible, but also highly beneficial. The continuation of this line of research promises to revolutionize smart and automated manufacturing in numerous industries globally.