Quality Control in

Manufacturing
• Quality control in manufacturing is a crucial process that
ensures products meet predefined standards and
specifications.
• It plays a vital role in delivering high-quality goods to
customers, minimizing production losses, and maintaining
brand reputation.
• Quality control methods are implemented at various
stages of production, including raw material inspection,
in-process checks, and final product evaluation.
• Effective quality control practices are essential for any
manufacturing organization, regardless of its size or
industry.
Common Quality Control Methods

1 Statistical Process Control (SPC) 2 Acceptance Sampling

SPC uses statistical techniques to monitor and Acceptance sampling involves inspecting a
control manufacturing processes, aiming to random sample of products from a batch to
identify and eliminate sources of variation that determine the quality of the entire batch. This
could lead to product defects. It involves collecting method is often used when it is impractical or
data, analyzing trends, and applying corrective cost-prohibitive to inspect every single product. It
actions to maintain consistent product quality. relies on statistical calculations to make decisions
about accepting or rejecting the batch.

3 Visual Inspection 4 Dimensional Inspection

Visual inspection is a basic method that relies on Dimensional inspection uses specialized
human observation to detect defects. It is often measuring tools, such as calipers, micrometers,
used for simple checks, such as verifying the and coordinate measuring machines (CMMs), to
color, shape, or size of products. However, it can verify the dimensions of products. This method
be subjective and prone to human error, especially ensures that products meet precise tolerances and
in situations where subtle defects are difficult to specifications, which are critical for their
see. functionality and performance.
 Literature
Sr.no Title
survey:
Author

1. Automated Defect Sunny Behal, The paper discusses automated defect detection in physical components using machine
Detection in Physical Japinder Singh learning. It highlights the transition from manual inspection, which is prone to errors, to
Components using machine learning approaches that improve detection speed and accuracy. The authors propose a
Machine Learning model based on the MobileNet-SSD architecture, optimized for resource efficiency, which
outperforms traditional deep learning methods in defect detection. The paper describes the
model's development, pre-processing techniques, and training on datasets. Results show higher
precision in detecting surface defects with minimal computational resources.

2. Automated defect Tobias Schon, The paper explores using neural networks to detect defects in X-ray images of light metal
recognition in X-ray Rasmus Hvingel by, castings. It compares training on simulated versus real-world data, concluding that a
projections using Bishwajit Goswami combination of both yields the best results. The closer simulated data matches real-world
neural networks conditions, the better the defect detection performance. This method aims to reduce the costs
trained on simulated of real-world data collection and setup.
and real-world data

3. Automated 3D Defect Dr. Frank Sukowski, The paper presents a method for automated 3D defect detection using simulated reference
Detection based on Daniel Suth, parts in industrial CT scans. It overcomes challenges like limited scan time and image artifacts by
Simulated Reference Adrian Waldyra comparing parts to simulated "flawless" models, enhancing defect detection accuracy. This
approach eliminates the need for real defect-free parts, improving reliability in noisy
environments, especially for automotive components
4. Automated Yi-Cheng Huang, The paper presents an Automated Machine Learning (AutoML) system for defect detection on
Machine Learning Kuo-Chun Hung, cylindrical metal surfaces using deep learning models like VGG-16, ResNet-50, and MobileNet. The
System for Defect Jun-Chang Lin study compares Transfer Learning (TL) and AutoML approaches. The AutoML model, combining
Detection on modules from these networks, achieved 95.5% accuracy, while AutoKeras, another AutoML tool,
Cylindrical Metal reached 99.83% accuracy. This system reduces manual effort and improves accuracy in defect detection
Surfaces for smart manufacturing.

5. Automated Fabric Henry Y.T Ngan, The paper provides a review of automated fabric defect detection methods, classifying them into motif-
Defect Detection Grantham Kwok based and non-motif-based approaches. It compares seven detection techniques, outlines their
Hung Pang strengths and weaknesses, and discusses future research directions.

6. Automatic Ahmed AlRatrout , This method measures in-situ contact angles between immiscible fluids using segmented pore-scale X-
measurement of Ali Q Raeini, ray images. It identifies and meshes fluid/fluid and fluid/solid interfaces, applies Gaussian smoothing,
contact angle in Branko Bijeljic, and tracks the three-phase contact line to compute contact angles using vector dot products. The
pore-space images Martin J Blunt technique is validated with synthetic images and applied to oil/brine systems under reservoir
conditions, producing highly accurate contact angle measurements. It offers a rapid, precise
characterization of pore-scale wettability, crucial for improving hydrocarbon recovery and carbon
dioxide storage.

7. Automated Shubham Kakirde, Efficient warehouse management is crucial for modern supply chains, especially with the rise of e-
Dimension Shubham Jain, shopping. Many warehouses still use manual labor for tasks like measuring object dimensions, which is
Measurement Swaraj Kaondal, slow and inefficient. In a post-Covid world, automated, contactless systems are becoming more
System Reena Sonkusare, important for safety and efficiency. A proposed solution involves using a single line laser and camera
Rita Das module to quickly and accurately measure object dimensions using 3D reconstruction. This method
balances accuracy, speed, and power consumption, making it well-suited for industrial applications.
8. Research in automated Ronie Navon Recent research in Construction Automation (CA) has shifted towards Automated Project
measurement of project Performance Control (APPC), which uses Automated Data Collection (ADC) technologies to
performance indicators monitor project performance indicators in real-time, such as cost, schedule, and productivity.
Traditional manual data collection methods are labor-intensive, error-prone, and slow, limiting
real-time control. Technologies like barcodes, RFID, GPS, video, and LADAR are being explored
to automate data collection, enabling more accurate and efficient project control. These
innovations aim to improve construction management by providing real-time feedback, helping
address deviations more effectively.

9. Camera based automatic Adam Mrozinski, Recent advancements in video capture technology have led to the development of automated
system for tool Adam Klodowski video-based systems for monitoring and analyzing construction tools and equipment. These
measurement and systems utilize high-quality video cameras to collect data in real-time, allowing for efficient
recognition analysis of operational metrics. By integrating image processing algorithms with user-friendly
software, these systems enhance the identification and performance assessment of tools. The
paper highlights the benefits of utilizing such technologies, emphasizing their practical
implications in improving monitoring processes and overall operational efficiency in
construction environments.

10. Automatic Seounggeun Kim , This paper proposes an automated garment size measurement method that combines deep
measurements of Haejun Moon , learning models with LiDAR point cloud data. By using a mobile device equipped with both a
garment sizes using Jaehoon Oh , camera and a LiDAR sensor, the approach captures 2D images and 3D spatial information
computer vision deep Yonghak Lee , simultaneously, enabling accurate distance calculations for clothing size measurements. The
learning models and Hyun Kwon, and method shows high precision, with average relative errors of 1.59% and 2.08% in various
point cloud data Sunghwan Kim environments. This innovative technique addresses the limitations of traditional methods,
streamlining the sizing process in the fashion industry while minimizing labor costs and
reducing sizing discrepancies for consumers.
Automated Defect
Detection Technique
1. Machine Vision Systems
2. Non-Destructive Testing (NDT)
3. Ultrasonic Testing (UT)
4. Vision Measurement Machine
 1. Machine Vision Systems:
Advantages :
• Accuracy: Consistently captures and processes visual data with high precision, reducing human error.

• Speed: Performs inspections and measurements faster than human operators.

• Non-contact Measurement: Allows measurements without physical contact, preventing damage to delicate components.

• Automation: Integrates easily into automated production lines, increasing productivity.

Disadvantages :
• Initial Cost: High setup cost due to advanced hardware and software.

• Complex Setup: Requires specialized expertise for configuration and maintenance.

• Environmental Sensitivity: Performance can be affected by lighting conditions, reflections, and other environmental factors.

• Limited Flexibility: May require adjustments or retraining when products or environments change significantly.
 2.Non-Destructive Testing (NDT):
Advantages :
• Preservation of Material: Tests are conducted without causing any damage to the tested components, allowing their
continued use.

• Early Detection: Identifies defects and flaws early in the production or operational stages, preventing failures.

• Cost Efficiency: Reduces repair and downtime costs by detecting issues early without interrupting operations.

• Safety Enhancement: Improves the safety of structures, machinery, and systems by ensuring integrity without destructive
processes.

Disadvantages :
• Complex Interpretation: Requires specialized knowledge and expertise to interpret results accurately.

• Initial Equipment Cost: High cost of advanced NDT equipment and training.

• Surface Limitations: Some NDT methods may be limited to detecting surface or near-surface defects, missing internal issues.

• Time-Consuming Setup: Proper preparation and calibration can be time-consuming depending on the method used.
 3.Ultrasonic Testing (UT):
Advantages :
• High Accuracy: Provides precise measurements of flaw size and location, enhancing reliability in inspections.

• Non-Destructive: Allows the inspection of materials without causing any damage to the object being tested.

• Immediate Results: Delivers real-time feedback, enabling quick analysis and decisions.

• Deep Penetration: Can detect internal defects deep within a material, making it highly effective for thick or dense objects.

Disadvantages :
• Requires Skilled Technicians: Accurate interpretation of results requires trained and experienced personnel.

• Surface Requirements: The surface must be well-prepared (clean and smooth) for accurate results, and coupling
agents are often needed.

• High Equipment Cost: Advanced ultrasonic testing equipment can be expensive to acquire and maintain.

• Limited for Complex Shapes: May struggle with irregularly shaped or highly attenuative materials, reducing its
effectiveness in certain applications.
4.Vision Measurement Machine:
Advantages :
• High Precision: Provides accurate measurements of small and intricate components, often at micron-level precision.

• Non-Contact Inspection: Measures objects without physical contact, avoiding damage or deformation of delicate parts.

• Automated Measurements: Increases efficiency by automating the measurement process, reducing human error.

• Versatility: Can measure various dimensions like length, width, height, and angles, making it suitable for complex shapes.

Disadvantages :
• Expensive Setup: High initial cost for equipment and setup, as well as maintenance.

• Requires Controlled Environment: Performance can be affected by lighting, vibration, or temperature fluctuations.

• Limited Depth Measurement: Some machines may struggle to measure deep or recessed features accurately.

• Skilled Operation Needed: Proper usage and interpretation of results require skilled operators, adding training costs.
VMM:
• High Cost: Expensive setup, maintenance, and calibration.
• Limited Flexibility: Difficult to adapt to new parts or defects.
• Slow Inspection Speed: May not keep up with high-speed production.
• Operator Dependency: Requires skilled personnel for operation and calibration.
• Limited Flexibility: Difficult to adapt to new parts or defects.
• Complex Setup: Lengthy setup and recalibration for new tasks.

ML:
• Cost-Effectiveness: Lower long-term costs compared to traditional systems.
• Flexibility: Adapts to various defect types and new patterns.
• Automation: Enables real-time, automated inspection.
• Continuous Improvement: Improves accuracy with more data.
• Complex Defect Detection: Identifies subtle and intricate defects.
• Versatility: Handles multiple defect types (surface, dimensional, etc.).
 Why we
•choose ML:
High Accuracy in Defect Detection: Machine Learning models (e.g., CNNs) are capable of learning complex
patterns, enabling precise identification of subtle defects that are difficult to detect manually.

• Automation and Efficiency: Image Processing techniques (e.g., filtering, edge detection) automate feature
extraction, speeding up defect analysis and reducing manual effort.

• Adaptability: ML models can be trained on new defect types, making the system adaptable to different scenarios
and components.

• Real-Time Analysis: Integration of ML and IP allows for rapid, real-time defect detection, providing immediate
feedback during production.

• Reduced Human Error: Automation minimizes human intervention, reducing inconsistencies and ensuring
consistent quality control.

• Scalability and Flexibility: The combined use of ML and IP can be scaled for different production lines, defect
types, and quality assurance needs.
 The Case for Machine Learning and Image
Processing in Quality Control
• Machine learning (ML) and image processing (IP) are emerging technologies that are revolutionizing
quality control in manufacturing.
• These technologies offer significant advantages over traditional methods by providing greater accuracy,
speed, and automation. ML algorithms can be trained to identify subtle defects that are difficult or
impossible for human inspectors to detect, while IP techniques can analyze images of products in real-time
to identify and classify defects.
• These capabilities allow for more effective quality control, resulting in higher product quality, reduced
production costs, and improved customer satisfaction.
Implementing ML and IP for Quality Control
Implementing ML and IP for quality control requires a systematic approach that involves several steps:

1 1. Data Collection
The first step is to collect a large dataset of images of products, including both good and defective examples. This dataset will be used to train the ML model.

2 2. Model Training
The ML model is trained on the collected dataset, learning to identify patterns and features that distinguish defective products from good ones. This
involves feeding the model images and labels indicating whether each image represents a defective product or not.

3 3. Model Evaluation
After training, the model is evaluated on a separate set of images to assess its accuracy and performance. This ensures that the model is reliable and can
accurately identify defects in real-world scenarios.

4 4. System Integration
The trained model is then integrated into the manufacturing process, where it can be used to inspect products in real-time. This integration typically
involves connecting cameras and sensors to the system, which capture images of products and feed them to the ML model.

5 5. Continuous Improvement
The performance of the ML and IP system is continuously monitored and improved. This involves collecting feedback from the system's performance,
retraining the model with new data, and refining the system's parameters to enhance its accuracy and effectiveness.
How ML and IP Systems Work for Quality Control
ML and IP systems work by analyzing images of products and identifying patterns that indicate defects. The process typically involves the following steps:

1. Image Acquisition
The system acquires images of products using cameras or sensors. These images can be captured in real-time as products move along a conveyor

2. Image Preprocessing
The acquired images are preprocessed to enhance their quality and remove noise. This step includes operations like resizing, cropping, and
converting the images to grayscale or a different color space.

3. Feature Extraction
Features that are relevant for defect detection are extracted from the preprocessed images. These features can include color variations,
shape deviations, texture patterns, and other characteristics that distinguish defective products from good ones.

4. Defect Classification
The extracted features are used to classify the products as either good or defective. This is done by comparing the features to patterns
learned during the model training process. The system can also categorize defects into different types, such as scratches, dents, or missing
parts.

5. Output and Action

The system outputs the classification results, highlighting any detected defects and providing information about their location and type.
This information can be used to trigger automated actions, such as rejecting defective products, adjusting manufacturing processes, or
alerting operators for further inspection.
The Process of ML and IP-Driven Quality Control
The process of ML and IP-driven quality control involves several steps, from image acquisition to defect identification and action. Here is a

1. Image Acquisition 1
Cameras or sensors are strategically placed on the
production line to capture images of products as they move
along the conveyor belt. These images can be captured at 2 2. Data Preprocessing
multiple angles and resolutions to ensure comprehensive The acquired images are preprocessed to remove any noise
inspection. or distortions that might interfere with defect detection.
3. Feature Extraction 3 This includes adjusting brightness, contrast, and color
The preprocessed images are analyzed to extract features balance, as well as performing image filtering and edge
that are relevant for defect detection. These features can enhancement.
include shape, size, color, texture, and other characteristics
that distinguish defective products from good ones. 4 4. Defect Classification
Different feature extraction techniques are available, The extracted features are fed into a trained machine
depending on the type of defects being targeted. learning model that classifies the products as either good or
defective. The model uses algorithms to identify patterns
5. Action 5 and anomalies in the features, based on its training data.
Based on the classification results, the system triggers The classification results can also indicate the type of
appropriate actions. Defective products can be defect detected.
automatically rejected and sent for rework or disposal.
Alternatively, the system can alert operators for manual
inspection or adjust manufacturing processes to
minimize future defects.
Optimizing Quality Control with ML and IP
ML and IP offer numerous advantages for optimizing quality control in manufacturing. By leveraging
these technologies, manufacturers can achieve significant improvements in product quality,
efficiency, and cost savings. Key takeaways include:
Increased Accuracy Improved Efficiency

ML and IP systems can detect subtle defects that are Automated defect detection allows for faster
difficult or impossible for human inspectors to see. inspection times, enabling manufacturers to process
This leads to higher product quality and reduced products more efficiently and reduce production
customer complaints. delays. This also frees up human inspectors to focus
on more complex tasks.

Reduced Costs Data-Driven Decision Making

By minimizing defects and reducing rework, ML and These systems collect valuable data that can be
IP systems can significantly reduce production costs. used to identify trends and patterns in product
They also lead to lower labor costs as fewer quality. This data can be used to improve
inspectors are required for manual inspection. manufacturing processes and prevent future defects.
Conclusi
on: enhancing quality control.
• Summary: Developed an effective automated defect detection system for ferrules,

• Key Findings: Achieved high accuracy and efficiency, outperforming manual inspection
methods.

• Production Impact: Reduced inspection time and costs while improving product quality.

• Future Work: Explore expanding system capabilities and integrating predictive

maintenance features.

• Final Thoughts: Demonstrates the transformative potential of automation and AI in

manufacturing quality control.