Predictive Maintenance in Smart Industry

This project encompasses the development of an Industrial Internet of Things (IIoT) solution for predictive maintenance (PdM) of machines using vibration analysis and motor current signature analysis. PdM techniques analyze the condition of the in-service industrial equipment to determine when maint

Project Title

Predictive Maintenance in Smart Industry

Project Area of Specialization

Internet of Things

Project Summary

This project encompasses the development of an Industrial Internet of Things (IIoT) solution for predictive maintenance (PdM) of machines using vibration analysis and motor current signature analysis. PdM techniques analyze the condition of the in-service industrial equipment to determine when maintenance is required. PdM can result in cost savings compared to time-based maintenance because we perform maintenance only when needed.

This project includes the development of a machine fault simulator, deploying sensors to collect real-time condition data, and a cloud architecture where the sensors transmit real-time data. It also involves the design of an algorithm responsible for the classification and identification of potential faults. Figure 1 shows the overall schematic of the project.

Predictive maintenance finds applications in supporting maintenance programs in many industries to help improve fault detection, diagnostics, and prognosis.

Figure 1: Overall schematic of the project

Project Objectives

This project is basically focused on:

• Design and development of a prototype machine for fault generation

This machine will generate various faults, including misalignment, unbalance, and looseness. Faulty bearings and gears with known faults can be installed to study all kinds of bearing and gear faults.

• Deployment of sensors

We shall use two vibration sensors and a current sensor in this project. The two vibration sensors will be installed on bearing housings and the current sensor on the power cable of the motor. These sensors will provide data for Vibration Signature Analysis and Current Signature Analysis.

• Data acquisition and transmission over cloud

The vibration and current sensors shall transmit data to a microcontroller. This microcontroller will then transmit the data in raw form or a preprocessed form to an on-premise cloud server, via WiFi. The server will will also allow remote access to authorized users.

• Condition Monitoring using vibration analysis and current signature analysis

The vibration signals and the current signal acquired by the sensors will then be processed in the cloud using several methods including, time domain analysis, frequency domain analysis, and time-frequency domain analysis.

• Development of classifier to identify developing faults

A machine learning algorithm will make predictions using the current and previously analyzed signals, stored in the server's database, to identify possible developing faults at an early stage. A feedback can be set to improve the algorithm's prediction accuracy overtime.

Project Implementation Method

First, we have manufactured a machine fault simulator with 1 motor, 2 bearings, 1 flywheel, and a load. We can change the bearing with a faulty bearing as well. Vibration and current sensors are attached to the machine, which are interfaced by a Raspbery Pi 3. The Raspberry Pi acts as the edge device and sends data to the cloud server in a periodic manner. We are using Zetta as our IoT cloud platform currently running on a local PC. Once the server receives the data, the first stage of digital signal processing starts.

Figure 2 shows the work flow strucure of the project.

Figure 2: Work flow structure of the project

1. Pre-processing: Pre-processing removes noise from the data and formats the data in a tabular form.
2. Extracting basic features: Once the data is ready, we extract several basic features like mean, standard deviation, variance, kurtosis, shape, and crest factor. 
3. General Classification: Before applying extensive signal processing to the processed data, we need some indicator if it is faulty. For this purpose, we use a machine learning model called Logistic Regression to identify if the data points are healthy or not healthy.
4. Fault Determination: This is the most important, complex, and as well as computationally expensive step in the whole process. If the above step classifies the data points are healthy, all other steps are skipped else, the data points are now subjected to various signal processing techniques like wavelet transformation and special kurtosis with demodulation. With the help of these techniques, features like mean, kurtosis, etc. are again calculated. Threshold values for faults in different parts are already calculated, parallel, and then the comparison gives us the answer to our questions like what the type of fault is and where it is present. The machine learning approach along with data-driven approach may be used in conjunction here to boost the accuracy of the results.
5. Dashboard: On the IoT server, a web-based application will be running, i.e., a dashboard, where a user can see all the information of its machine, its faults, its features data, etc. If a fault is detected a notification alert will be generated and sent to the system administrator and the technical team as well with complete information as to where, and which fault is present

Benefits of the Project

Failure of any equipment in industry can be catastrophic and results in losses, if not predicted earlier. If the failure occurs at the time of maximum production, it can result in the loss of all the material currently in process and a substantial capital loss. Such failures can be avoided by detecting the current state of equipment beforehand. The beforehand detection results in a better prognosis of the particular fault and helps to ascertain whether the equipment is operating within safe limits or not. It is crucial to understand that predictive analytics is a journey and not a destination. It starts with identifying the right set of data points, integrating with the machine to ingest real-time data, and improving the data quality through live tracking of machine failures. Data preparation and data quality are the essential inputs for any predictive model. More high-quality data we can feed into the predictive model, the better its accuracy.

The benefits of this project include:

• A practical example of a real-time condition monitoring system interfaced through IIoT 
• Allows manufacturers to lower maintenance costs
• Extends equipment life
• Reduces downtime
• Improves production quality by addressing the problem beforehand
• Prevents cascading failure

Technical Details of Final Deliverable

The technical details of this project covers the following secions:

• Prototype Machine Fault Simulator

A prototype machine for fault generation is designed. Since most industrial machines consists of bearings, induction motors, shafts, and other components, we can use this fault simulator to study common machinery faults without compromising production schedules or profits of the industry. Furthermore, it allows us to introduce different types of faults in a controlled environment. In this way, the behavior, and faults of common industrial machines, can be properly simulated, without having to go there. Figure 3 shows a prototype fault generation machine.

Figure 3: Prototype fault simulator machine

• Control box design

A control box is designed to provide power to the sensors and motor. Figure 4 shows the design of the control box. It contains a variable frequency drive (VFD) for controlling the speed of the motor. Beside the VFD is a breaker, and the microcontroller is placed inside the control box. There are connectors in between. All the equipment is mounted on DIN rails.

Figure 4: Control box 

• UI interface

A user dashboard, hosted on an IoT server will display the health of the machine. Any prediction made by the classifier of a developing fault will be notified to the user. Authorized users may also be able to access previous data regarding the health of the machine stored on the server's database, via this dashboard. The UI will be set up using a web app to allow access from any device.

• Fault detection techniques:

After acquiring data from the sensors successfully, features are extracted to distinguish different signals that are generated by different types of faults. The following techniques are used for feature extraction:

1. Basic time domain techniques-Average, Mean, RMS, Peak Amplitude, etc.
2. Frequency spectrum statistical features-AM, GM, RMS
3. Time-Frequency Domain analysis-STFT, Wavelets, Spectral Kurtosis, Demodulation

After feature extraction, signals are transferred to the cloud for further processing. The next step of processing is feature selection. Among all the features, we will see which feature is giving us most useful information. We are using SVM technique for this purpose. 

As a last step of predictive maintenance, we are applying machine learning techniques to separate out the faulty and healthy data, predicting the severity of fault and displaying it on the screen.

Final Deliverable of the Project

HW/SW integrated system

Core Industry

Manufacturing

Other Industries

Core Technology

Others

Other Technologies

Internet of Things (IoT)

Sustainable Development Goals

Industry, Innovation and Infrastructure, Responsible Consumption and Production

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com