Page 39 - IJAMD-1-2

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

Table 4. Advantages and limitations of various artificial intelligence techniques
AI techniques Aspects Explanation
ML Overview ML encompasses a broad range of algorithms that enable systems to learn patterns and make predictions from
data. In AM, ML techniques are used for defect detection, process optimization, and material property prediction.
Advantages • Predictive capabilities: ML models can predict potential defects and optimize parameters based on historical data,
improving process reliability.
• Versatility: ML algorithms, such as decision trees, random forests, and SVMs, can be applied to various types of
data, including numerical, categorical, and textual data.
• Scalability: Many ML techniques are scalable and can be applied to large datasets, making them suitable for
industrial applications.
Limitations • Data dependency: ML models require large amounts of high‑quality data for training, which can be difficult to
obtain in some AM environments.
• Overfitting: ML models, especially complex ones, can overfit the training data, leading to poor generalization of
new data.
• Interpretability: Some ML models, particularly ensemble methods, can be difficult to interpret, making it
challenging to understand the decision-making process.
Practical • SVMs for material classification: Used to classify material properties based on process parameters and sensor
applications data, providing reliable quality assurance.
• Random forests for defect prediction: Applied to predict defects based on historical process data, enabling
proactive quality control.
Computer Overview CV involves the use of algorithms to process and analyze visual data, such as images and videos. In AM, CV
vision techniques are used for real-time monitoring and defect detection.
Advantages • Real‑time analysis: CV systems can analyze visual data in real‑time, allowing for immediate detection and
correction of defects.
• High accuracy: Advanced CV techniques can achieve high accuracy in identifying defects and surface anomalies,
improving overall quality assurance.
• Automation: CV systems can automate the inspection process, reducing the need for manual intervention and
increasing efficiency.
Limitations • Complexity: Implementing CV systems can be complex and require specialized knowledge in image processing
and analysis.
• Hardware requirements: CV systems often require high‑resolution cameras and significant computational
resources, which can be costly.
• Sensitivity to conditions: Performance can be affected by variations in lighting, camera quality, and
environmental conditions.
Practical • Defect detection in SLM: CV systems are used to monitor layer‑by‑layer images in SLM, identifying defects such
applications as porosity and surface roughness in real time.
• Dimensional accuracy monitoring: CV techniques are employed to ensure parts are produced within specified
dimensional tolerances, enhancing quality control.
Neural Overview Neural networks, particularly deep learning models like CNNs, are a subset of ML techniques that mimic the
networks human brain’s structure and function. They are highly effective for complex pattern recognition tasks.
Advantages • High performance: Neural networks can achieve state‑of‑the‑art performance in tasks such as image recognition,
defect detection, and process optimization.
• Automated feature extraction: Unlike traditional ML methods, neural networks can automatically learn relevant
features from raw data, reducing the need for manual feature engineering.
• Flexibility: Neural networks can be adapted to a wide range of applications and data types, making them versatile
tools for quality assurance.
Limitations • Data‑intensive: Training neural networks requires large amounts of labeled data, which can be a significant
barrier in AM applications.
• Computationally expensive: Neural networks demand substantial computational resources, often necessitating
specialized hardware like GPUs.
• Black‑box nature: The decision‑making process of neural networks is often opaque, posing challenges for
interpretability and trust in critical applications.
Practical • CNNs for defect detection: Applied to analyze in‑situ monitoring images and identify defects in real‑time,
applications significantly improving defect detection accuracy in AM processes.
• RNNs for process optimization: Used for sequential data analysis, RNNs can optimize AM process parameters by
learning from historical process data.
Abbreviations: AM: Additive manufacturing; ML: Machine learning; RNN: Recurrent neural network; CNN: Convolutional neural network;
SLM: Selective laser melting; CV: Computer vision.

Volume 1 Issue 2 (2024) 33 doi: 10.36922/ijamd.3455