Page 28 - IJAMD-1-2

International Journal of AI for
Materials and Design
AI-driven quality assurance in AM

or impractical to achieve using traditional manufacturing AI in AM quality assurance involves several steps, from
techniques. From intricate lattice structures in biomedical data collection to model deployment. Here is a detailed
implants to lightweight aerospace components with breakdown of how AI techniques are applied:
optimized internal geometries, the versatility of AM has (i) Data collection
fueled innovation and pushed the boundaries of design • Sensors and monitoring systems: AM processes
and engineering. are equipped with sensors and monitoring
However, despite its transformative potential, AM poses systems that capture real-time data, including
unique challenges, particularly in the realm of quality temperature, pressure, laser power, and visual
assurance. Traditional manufacturing processes typically data (images or videos) of the build process.
involve well-established quality control measures, such • Data types: The collected data include numerical
as in-process inspections and post-production testing, process parameters, time-series data, and image
to ensure product quality and reliability. In contrast, data of each layer being printed.
2,3
the dynamic and additive nature of AM introduces new (ii) Data preprocessing
complexities and uncertainties that traditional quality • Cleaning and normalization: Raw data are
assurance methods may struggle to address effectively. cleaned to remove noise and inconsistencies, and
One of the primary challenges in AM quality assurance subsequently normalized to ensure uniformity.
is the inherent variability in material properties and process • Feature extraction: Relevant features are extracted
parameters. Unlike conventional manufacturing methods, from the raw data. For instance, in image data,
where material properties are relatively uniform and features might include texture, edges, and shapes
predictable, AM processes often rely on powders, resins, or indicative of defects.
4
filaments with varying compositions and characteristics. In (iii) Model training
addition, factors such as build orientation, laser power, and • Supervised learning: In supervised learning,
printing speed can significantly influence the mechanical labeled datasets are used to train models to
properties and structural integrity of AM parts, making it recognize patterns associated with defects
challenging to establish consistent quality standards. or optimal process conditions. Common
4,5
Moreover, the layer-by-layer nature of AM introduces algorithms include decision trees, support vector
the potential for defects and anomalies at each stage of the machines (SVMs), and neural networks.
printing process. Common defects in AM parts include • Unsupervised learning: Techniques such as
clustering and anomaly detection are used to
porosity, warping, delamination, and surface irregularities, identify unusual patterns or outliers that may
which can compromise the mechanical strength, indicate defects without prior labeling.
dimensional accuracy, and surface finish of the final
product. Detecting and mitigating these defects require (iv) Model deployment
5
robust quality assurance strategies capable of monitoring • Real-time monitoring: Trained models are
and controlling the entire manufacturing process in real deployed to monitor the AM process in real time,
time. providing immediate feedback on potential defects
or deviations from optimal process parameters.
In recent years, artificial intelligence (AI) has emerged • Process optimization: AI models can dynamically
as a promising tool for addressing the challenges of quality adjust process parameters to optimize the
assurance in AM processes. AI algorithms, including manufacturing process, ensuring consistent
machine learning (ML) and deep learning techniques, quality and reducing waste.
can analyze large volumes of data generated during the
printing process, including sensor readings, imaging (v) Feedback and improvement
data, and CAD/computer-aided manufacturing models, • Continuous learning: AI systems can continuously
to identify patterns, predict defects, and optimize process learn from new data, improving their accuracy
parameters. By harnessing the power of AI, manufacturers and adaptability over time.
6
can enhance the quality, reliability, and efficiency of AM • Human-in-the-loop: Human experts can review
processes, paving the way for widespread adoption AI predictions and provide feedback, enhancing
across industries. This section explains how AI works in the system’s performance and reliability.
the context of AM quality assurance and highlights the Numerous research efforts and practical
work leveraging AI to improve the quality, reliability, and implementations have demonstrated the potential of AI in
efficiency of AM processes. 7 enhancing AM quality assurance. Key examples include:

Volume 1 Issue 2 (2024) 22 doi: 10.36922/ijamd.3455