ABSTRACT:
Loose fasteners are a common cause of mechanical failure in industrial and structural systems, leading to reduced efficiency, increased maintenance costs, and potential safety hazards. Early detection of screw looseness is therefore critical for ensuring system reliability and preventing unexpected breakdowns. This project presents the design and implementation of an intelligent, IoT-enabled loose screw detection system that combines sensor-based machine learning with vision-based deep learning for accurate and real-time analysis.
The proposed system utilizes two screws instrumented with MPU6050 accelerometer sensors and vibration sensors to continuously monitor motion and vibration characteristics. These signals are collected and processed to extract meaningful features such as average acceleration and vibration patterns over a fixed duration. A Random Forest classification model is then employed to analyze these features and determine whether each screw is in a tight or loose condition. This sensor-based approach effectively captures the physical behavior of the screws under dynamic conditions.
In addition to mechanical sensing, the system incorporates a Convolutional Neural Network (CNN) model to perform visual inspection of screw threads using images captured through a Raspberry Pi camera. The CNN model classifies the thread condition into three distinct categories: fully tight (0% loose), partially loose (50% loose), and fully loose (100% loose). This vision-based analysis provides a more detailed understanding of the degree of looseness, complementing the binary output of the sensor-based model.
To enhance usability, a graphical user interface (GUI) is developed using Tkinter, allowing users to initiate scans, monitor system status, and view results in real time. Furthermore, the system is integrated with a web-based platform via an API, enabling continuous transmission of sensor data and prediction results to a remote server. This facilitates real-time monitoring, data logging, and potential future implementation of alert mechanisms for predictive maintenance.
By combining multiple sensing modalities and intelligent algorithms, the proposed system offers a reliable, cost-effective, and scalable solution for loose screw detection. It has significant applications in industrial automation, structural health monitoring, and maintenance systems, where early fault detection is essential for ensuring safety and operational efficiency.

INTRODUCTION:
In modern industrial and mechanical systems, the reliability and safety of structures heavily depend on the integrity of fastening components such as screws and bolts. Even a slight loosening of these fasteners can lead to severe consequences, including mechanical failure, system inefficiency, increased maintenance costs, and potential safety hazards. Loose screws are particularly problematic in applications involving continuous vibration, dynamic loading, or harsh environmental conditions, where they may gradually lose their tightness over time. Therefore, the timely detection of screw looseness is a critical requirement in predictive maintenance and condition monitoring systems.
Traditional methods for detecting loose screws primarily rely on manual inspection or periodic maintenance routines. These approaches are often time-consuming, labor-intensive, and prone to human error, especially in large-scale industrial environments. Moreover, manual inspection cannot provide real-time monitoring, which increases the risk of undetected faults leading to unexpected system failures. With the advancement of sensing technologies and artificial intelligence, there is a growing need for automated, accurate, and real-time solutions that can continuously monitor the condition of fasteners without human intervention.
This project addresses these challenges by proposing an intelligent loose screw detection system that integrates sensor-based machine learning with vision-based deep learning techniques. The system is designed using a Raspberry Pi platform, making it compact, cost-effective, and suitable for real-time embedded applications. Two screws are instrumented with MPU6050 accelerometer sensors and vibration sensors to capture motion and vibration characteristics associated with their physical state. These signals are processed and analyzed using a Random Forest machine learning model to classify whether a screw is tight or loose based on its dynamic behavior.
In addition to sensor-based analysis, the system incorporates a Convolutional Neural Network (CNN) to perform visual inspection of screw threads using images captured by a Raspberry Pi camera. This enables the classification of thread conditions into multiple levels of looseness, providing a more detailed and accurate assessment compared to traditional binary detection methods. The combination of these two approaches enhances the robustness and reliability of the system by validating results through both physical and visual data.
To improve user interaction and system control, a graphical user interface (GUI) is developed using Tkinter, allowing users to initiate scans and view results in real time. Furthermore, the system is integrated with a web-based platform through an API, enabling remote monitoring and data logging. This IoT-based capability allows maintenance personnel to access system status from anywhere, facilitating faster decision-making and proactive maintenance strategies.
Overall, this project aims to develop a smart, automated, and scalable solution for loose screw detection by leveraging the capabilities of machine learning, deep learning, and IoT technologies. The proposed system not only improves detection accuracy but also contributes to enhancing safety, reducing downtime, and optimizing maintenance processes in industrial applications.

OBJECTIVES:
The primary objective of this project is to design and develop an intelligent and automated system for detecting loose screws using a combination of sensor-based machine learning and vision-based deep learning techniques. The system aims to provide accurate, real-time monitoring and analysis of screw conditions, thereby improving reliability and reducing the risk of mechanical failures. The detailed objectives of the project are as follows:
1. To develop a sensor-based detection mechanism
To design and implement a hardware setup using MPU6050 accelerometer sensors and vibration sensors mounted on screws to capture motion and vibration characteristics. The objective is to acquire reliable real-time data that reflects the physical condition of the screws under different states such as tight and loose.
2. To implement machine learning for screw condition classification
To utilize a Random Forest algorithm for analyzing sensor data and classifying the screw condition as either tight or loose. This involves preprocessing the collected data, extracting meaningful features, and training the model to achieve accurate prediction based on variations in acceleration and vibration patterns.
3. To design a vision-based thread analysis system
To develop a Convolutional Neural Network (CNN) model capable of analyzing images captured from a Raspberry Pi camera to detect the level of thread looseness. The objective is to classify the screw condition into three categories: fully tight, partially loose, and fully loose, providing a more detailed assessment compared to binary classification.
4. To integrate sensor and vision-based approaches
To combine both machine learning and deep learning techniques into a unified system that enhances detection accuracy and reliability. The objective is to leverage both physical sensing and visual inspection for cross-verification of results.
5. To develop a real-time user interface
To create a graphical user interface (GUI) using Tkinter that allows users to control the system, initiate scanning processes, and visualize results in real time. The interface should be user-friendly and suitable for practical deployment.
6. To implement IoT-based data transmission
To establish communication between the Raspberry Pi system and a remote web server using an API. The objective is to transmit sensor readings and prediction results for remote monitoring, data logging, and further analysis.
7. To enable real-time monitoring and analysis
To ensure that the system operates in real time, providing immediate feedback on screw conditions. This helps in early detection of faults and supports predictive maintenance strategies.
8. To design a cost-effective and scalable solution
To develop a system that is affordable, compact, and scalable for use in industrial and mechanical applications. The objective is to make the solution practical for real-world deployment without requiring expensive infrastructure.

9. To improve safety and reduce maintenance costs
To minimize the risks associated with loose screws by enabling early detection and timely intervention. This objective focuses on enhancing operational safety, reducing downtime, and lowering maintenance expenses.

Block Diagram