CE351 - Microcontrollers 2022 Spring
ESP32 Anomaly Detection Project
Name: Sean Eaton
Email: smeaton@fortlewis.edu

ESP32 Anomaly Detection Project

Introduction

The purpose of this project is to replicate the Edge AI Anomaly Detection project by Shawn Hymel which can be found on the Digikey YouTube channel here. Anomaly detection is a technique of detecting abnormalities or changes that may have severe negative effects depending on the system. In a piece of machinery an anomaly may lead to thousands of dollars of damage to the machine itself or to the things being manufactured such as in the case of a robotic arm tasked with a role in vehicle manufacturing. An anomaly causing the robotic arm to make mistakes in its movements could break the robotic arm or cause damage to the vehicle. In part I Shawn Hymel details how data can be collected using an ESP32 development board, an accelerometer, and a remote server on a local network.

Task 1:

For the first task the ESP32 needed to run a program capable of connecting to a server, reading accelerometer data, and transmitting that data to the server. Shawn Hymel provided an example Arduino sketch that was able to be used with some modifications. Hymel is using a different ESP32 development board and a MSA301 3-axis accelerometer breakout board which required some code adjustments in order to allow the usage of the MPU6050 6-axis gyroscope and accelerometer breakout boards available to us. The network SSID, password, and hostname IP addresses also had to be defined in the Arduino sketch. The hostname IP address can be found by running the 'ipconfig' command in a command prompt on a Windows PC. The IPv4 Address is the IP address needed for the Arduino sketch.
Instead of using the Adafruit_MSA301 and Adafruit_Sensor libraries I opted to use the MPU6050_tockn library that I have had success in using for other projects. The relevants parts of the code reading the MSA301 sensor were then replaced with code reading the MPU6050. There is no built in LED on the ESP32 DevkitC V4 development boards we have been using so discrete LED was used instead. The only additional library that was needed was the ArduinoJson library. The circuit on the breadboard can be seen below in Figure 1. The SCL and SDA pins of the MPU6050 were connected to the SCL and SDA pins on the ESP32 development board. These were GPIO pins 22 and 21 respectively. The power bank is a budget one that is able to provide 4700mAh at 5V.


Figure 1. Picture of the circuit implemented on breadboard.

Task 2:

The second task of this project consisted of running a simple server on my local network so that the ESP32 is able to transmit MPU6050 accelerometer readings to it. After transmitting the set of 200 samples the data is saved as a csv file in the specified folder. This is done using a Python script also provided by Hymel. There weren't any modifications to the Python script that were necessary for the code to run although the provided command needed the folder name, port, and the time for running the server defined. This command is shown below:
 
python http_accel_server.py -d <DIRECTORY> -p <PORT> -t <TIME TO RUN SERVER>
The port used in this tutorial is 1337, which is defined when running the Python script. The Arduino sketch already has the port defined as 1337 but it could be changed as long as the Arduino sketch is also changed. Visual Studio Code was used to execute the Python script. The relevant folder needs to be opened in Visual Studio Code and then the Run Python File button can be clicked on. It is after clicking on this button that an embedded terminal is opened where the command can be entered.

Results:
After getting the server and ESP32 to connect together the ESP32, MPU6050, and power bank were affixed to a ceiling fan I had access to using some tape. Figure 2 below shows how this was accomplished. Power hasn't been connnected to the ESP32 just yet,


Figure 2. The ESP32 and MPU6050 circuit taped to a ceiling fan.

After this the server was started on my PC. It was important to connect to my 2.4 GHz Wifi connection because the ESP32 is only able to connect to connections of that frequency (the 5 GHz connection is not supported). The server was started first because I thought that it would be more appropriate for data collection because the MPU6050 needed to run through a calibration process beforehand. This turned out to be unoptimal however since the first 10 files saved did not consists of accelerometer measurements when the fan was running. I started the server then plugged the ESP32 into power. After the ESP32 turned on the ceiling fan then needed to be turned on, resulting in the first 10 csv files not being the correct type of data we are interested in saving. An example of the feedback generated by the Python script running the server can be seen below in Figure 3. You can see that the csv file being created is within the 'fastest_speed_01' directory which denotes the fan speed and dataset number. 535 total csv files were created using the default 2400 seconds as the time to run the server. For each csv file 200 accelerometer samples are recorded. These are the accelerometer values on X, Y, and Z axes. An example of a CSV file can be seen in Figure 4.


Figure 3. Server script feedback from the Visual Studio Code embedded terminal.


Figure 4. An example of the acclerometer data being recored to each CSV file.

While the ESP32 is connected to the server and transmitting data the LED is set to blink so that it can be easily seen if the ESP32 is stil transmitting data. After the server run time is expired the LED ceases to blink and the server automatically shuts down. An image of the ESP32 during the transmission process can be seen below in Figure 5. It is important to note that the Python script running the server is not appropriate for use beyond testing on your own personal network. There is a lack of security features in the Python code so it is best that it was limited to my home for a short period.


Figure 5. The ESP32 blinks the LED on and off to indicate when data is being transmitted to the Python server.

Discussion

Although this project had the goal of detecting anomalies this wasn't able to be accomplished in time. For now the data collection process is able to be conducted and for future steps a large amount of data is needed for robust anomaly detection using machine learning. For future work on this project I would first recommend that one becomes a little familiar with using Visual Studio Code for Python development to make the process of getting the server up and running quickly. The next step of Shawn Hymel's tutorial delves into using Tiny Machine Learning (TinyML) to accomplish anomaly detection. Large amounts of data will be needed for feature extraction and for training the machine learning model to detect anomalies. I enjoyed learning about how a  remote data collection system can be created using the ESP32 and a WiFi connection. It was nice being able to run the sketch on the ESP32, run the Python script on my PC, and then wait for data collection to be completed automatically without any input from myself. I had previously done data collection for a Junior design course so that a pedestrian detection neural network could be trained using images from the FLC webcam. In that project the data collection process was pretty tedious.