You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
@@ -64,28 +64,29 @@ The goal of this application note is to demonstrate a motor anomaly detection an
64
64
- The [Arduino Nicla Sense ME Firmware](assets/Nicla_Opta_ML_Vibration_Anomaly_Detection.zip) to capture the data and send to the Opta™ via Bluetooth® Low Energy.
65
65
- The [Arduino Opta™ Firmware](assets/Nicla_Opta_ML_Vibration_Anomaly_Detection.zip) to capture the vibration data sent by the Nicla Sense ME via Bluetooth® Low Energy and transfer it to the Edge Impulse platform.
66
66
- The [Opta™ Deployment Firmware](assets/Nicla_Opta_ML_Vibration_Anomaly_Detection.zip) to enables real-time monitoring of the motor's condition.
67
-
67
+
- The [Edge Impulse public project](https://studio.edgeimpulse.com/public/580971/live).
68
+
68
69
***The model performance can be affected if the application is implemented on a very different environment than the one used for training. It's recommended to feed the dataset with new samples and retrain the model for a new and upgraded deployment.***
69
70
70
-
###Machine Learning Model for Motor Anomaly Detection
71
+
## Machine Learning Model for Motor Anomaly Detection
71
72
72
73
Machine Learning is a programming approach that enables devices to analyze raw sensor data, extract key features, and use them to recognize patterns or make predictions based on prior training. In this application note, machine learning is applied to identify motor anomalies by analyzing vibration data collected through the Nicla Sense ME.
73
74
74
75
To identify potential issues with the motor, a machine learning model was trained using the Edge Impulse platform. This model analyzes vibration patterns from the motor and predicts whether the motor is operating normally or exhibiting signs of anomalies.
75
76
77
+
78
+
76
79
First, the Nicla Sense ME was flashed with the [Arduino Nicla Sense ME Firmware](assets/Nicla_Opta_ML_Vibration_Anomaly_Detection.zip) that collects and transmits vibration data to the Opta™ via Bluetooth® Low Energy. Similarly, the Opta™ was flashed with the [Arduino Opta™ Firmware](assets/Nicla_Opta_ML_Vibration_Anomaly_Detection.zip) designed to receive the vibration data from the Nicla Sense ME and forward it to the Edge Impulse platform for further processing.
77
80
78
81
To transmit the data collected on the Opta™, it is necessary to use the Edge Impulse CLI, which must have been previously installed and configured. Using the Data Forwarder, it is possible to send the data to the Edge Impulse platform. To learn how to use the Data Forwarder, click [here](https://docs.edgeimpulse.com/docs/tools/edge-impulse-cli/cli-data-forwarder).
79
82
80
83
Once the data was collected, we used Edge Impulse to design a machine learning model capable of identifying anomalies in motor vibrations. The trained model is then deployed back to the Opta™ to monitor the motor health in real-time and detect anomalies as they occur.
81
84
82
-
Here is our model design:
83
-
84
-
85
+
### Model design:
85
86
86
87
As the Nicla Sense ME communicates with the Opta™ via Bluetooth BLE, the following characteristics can be applied as an example to ensure compatibility with the onboard Bosch sensors:
87
88
88
-
In the time series data block:
89
+
**In the time series data block:**
89
90
90
91
91
92
@@ -95,13 +96,13 @@ In the time series data block:
95
96
96
97
**Time series data** is critical in this application as it captures vibration measurements over consistent time intervals. The vibration data is composed of readings from the accelerometer on the Nicla Sense ME, corresponding to the X, Y, and Z axes. By analyzing these three-dimensional vibration values, the machine learning model can more accurately detect how vibrations evolve over time, identify patterns, and pinpoint deviations indicative of motor anomalies.
**Spectral Analysis** plays a crucial role in this application as it transforms raw vibration data into the frequency domain, uncovering patterns or anomalies that are not readily apparent in time-series data. By highlighting unique frequency components, this process significantly enhances the model's ability to detect motor issues, such as imbalance or misalignment, with improved precision.
103
104
104
-
In the learning block:
105
+
**In the learning block:**
105
106
106
107
107
108
@@ -123,7 +124,7 @@ Here is our Anomaly Explorer:
123
124
124
125
The "Anomaly Explorer" shows training data based on two features: "accX RMS" (Root Mean Square of acceleration along the X-axis) and "accY RMS" (along the Y-axis), which are related to vibration signals. The blue dots represent the training data, showing a central clustering pattern, while the two concentric ellipses define confidence regions: the inner ellipse encompasses most normal data, and the outer ellipse marks the threshold for anomalies. Data outside the outer ellipse would be flagged as anomalous, helping identify deviations from normal behavior. Currently, no test data is displayed, as indicated by the "No test data" selection.
125
126
126
-
> **Note**: It is important to note that the model’s performance may vary if implemented in an environment different from the one used for training. To ensure consistent results, it is recommended to gather new datasets from the specific deployment environment, retrain the model, and redeploy it for better adaptability.
127
+
***It is important to note that the model’s performance may vary if implemented in an environment different from the one used for training. To ensure consistent results, it is recommended to gather new datasets from the specific deployment environment, retrain the model, and redeploy it for better adaptability.***
127
128
128
129
For redeployment, use the Edge Impulse uploader to replace the existing model with the updated version, ensuring the system uses the most current model for anomaly detection.
129
130
@@ -149,10 +150,10 @@ Using the trained machine learning model, a code is generated to embed the model
149
150
150
151
### Nicla Sense ME Code
151
152
152
-
153
-
154
153
This is the initial code required for the setup. It runs on the Nicla Sense ME, configuring it to function as a Bluetooth Low Energy (BLE) peripheral. The code reads real-time accelerometer data from the X, Y, and Z axes and transmits it wirelessly to the Opta™, which operates as the BLE central device.
155
154
155
+
156
+
156
157
In this application note, key code sections will be explored to ensure the motor anomaly detection system becomes fully operational. The discussion will commence with a review of the necessary libraries:
157
158
158
159
- The `ArduinoBLE.h` enables Bluetooth® Low Energy (BLE) communication, allowing the Nicla Sense ME to transmit accelerometer data wirelessly to the Opta™. You can install this library by searching for it in the Library Manager within your Arduino IDE. For more details and documentation, visit the official [ArduinoBLE library documentation](https://docs.arduino.cc/libraries/arduinoble/).
@@ -161,9 +162,9 @@ In this application note, key code sections will be explored to ensure the motor
161
162
162
163
The Bluetooth® Low Energy (BLE) services and characteristics are specifically configured to support the features of this application. The primary service is designed to handle the transmission of accelerometer data for monitoring motor anomalies, making it ideal for real-time vibration analysis. A characteristic is defined to send the accelerometer's X, Y, and Z-axis data, which correspond to the motor's vibration levels. These three data points are combined and transmitted together in a single buffer. Both the service and characteristic are assigned unique and standardized BLE UUIDs to ensure seamless interaction between the Nicla Sense ME and the Opta™.
163
164
164
-
The code below represents the first part of the Nicla Sense ME code, and it is responsible for sets up the Nicla Sense ME for Bluetooth® Low Energy (BLE) communication to transmit accelerometer data (X, Y, and Z axes) as a single 6-byte buffer (`bufferSize`). It uses the `ArduinoBLE` and `Arduino_BHY2` libraries to initialize BLE functionality and access the accelerometer sensor. A BLE service (`niclaService`) and characteristic (`sensorCharacteristic`) are defined with unique UUIDs, allowing the central device (Opta™) to recognize and read vibration data efficiently. Data is sent every 100 milliseconds.
165
+
The code below represents the first part of the Nicla Sense ME code, and it is responsible for setting up the Nicla Sense ME Bluetooth® Low Energy (BLE) communication to transmit accelerometer data (X, Y, and Z axes) as a single 6-byte buffer (`bufferSize`). It uses the `ArduinoBLE` and `Arduino_BHY2` libraries to initialize BLE functionality and access the accelerometer sensor. A BLE service (`niclaService`) and characteristic (`sensorCharacteristic`) are defined with unique UUIDs, allowing the central device (Opta™) to recognize and read vibration data efficiently. Data is sent every 100 milliseconds.
165
166
166
-
The code can be downloaded [here](assets/Nicla_Opta_ML_Vibration_Anomaly_Detection.zip).
167
+
The full code can be downloaded [here](assets/Nicla_Opta_ML_Vibration_Anomaly_Detection.zip).
167
168
168
169
```arduino
169
170
#include <ArduinoBLE.h> // Include the ArduinoBLE library for Bluetooth functionality
@@ -276,7 +277,7 @@ void loop() {
276
277
// Copy accelerometer data into the buffer
277
278
memcpy(data, &accX, sizeof(accX)); // Copy X-axis data
278
279
memcpy(data + 2, &accY, sizeof(accY)); // Copy Y-axis data
279
-
memcpy(data + 4, &accZ, sizeof(accZ)); // Copy Z-axis da
280
+
memcpy(data + 4, &accZ, sizeof(accZ)); // Copy Z-axis data
280
281
281
282
// Write the buffer data to the BLE characteristic
This code demonstrates the configuration of the Opta™ as a central Bluetooth Low Energy (BLE) device, designed to connect to the Nicla Sense ME, which functions as a BLE peripheral. The program scans for BLE devices advertising a specific service UUID, connects to the first matching peripheral, and retrieves accelerometer data from its X, Y, and Z axes. The data is transmitted in real-time using a BLE characteristic.
299
298
299
+
300
+
300
301
This code is a critical part of a motor anomaly detection system, enabling the Opta™ to receive vibration data wirelessly for further processing in the Edge Impulse Platform.
301
302
302
-
-Similar to its use in the Nicla Sense ME code, the`ArduinoBLE.h` library will be utilized to implement Bluetooth® Low Energy (BLE) communication. In this code, it enables the Opta™ to function as a BLE central device, capable of discovering and connecting to BLE peripherals like the Nicla Sense ME. To install this library, search for `ArduinoBLE` in the Arduino IDE Library Manager. For more information and detailed documentation, visit the official [ArduinoBLE library documentation](https://docs.arduino.cc/libraries/arduinoble/).
303
+
-The`ArduinoBLE.h` library will be utilized to implement Bluetooth® Low Energy (BLE) communication. In this code, it enables the Opta™ to function as a BLE central device, capable of discovering and connecting to BLE peripherals like the Nicla Sense ME. To install this library, search for `ArduinoBLE` in the Arduino IDE Library Manager. For more information and detailed documentation, visit the official [ArduinoBLE library documentation](https://docs.arduino.cc/libraries/arduinoble/).
303
304
304
305
The code is configured to scan for a BLE peripheral that advertises a specific service UUID. Once connected, the Opta™ accesses a characteristic that provides accelerometer data from the peripheral. The accelerometer's X, Y, and Z data points are transmitted together in a single 6-byte buffer.
305
306
@@ -330,9 +331,11 @@ void setup() {
330
331
}
331
332
```
332
333
333
-
The second part enable the Opta™ to function as a BLE central device for real-time data acquisition. The code begins by scanning for BLE peripherals advertising a specific service UUID, identifying compatible devices, such as the Nicla Sense ME. Once a peripheral is discovered, the Opta™ attempts to connect, retrieves its attributes, and accesses the defined BLE characteristic containing accelerometer data.
334
+
The second part enables the Opta™ to function as a **BLE central device** for real-time data acquisition. The code begins by scanning for BLE peripherals advertising a specific service UUID, identifying compatible devices, such as the Nicla Sense ME. Once a peripheral is discovered, the Opta™ attempts to connect, retrieves its attributes, and accesses the defined BLE characteristic containing accelerometer data.
335
+
336
+
After establishing a connection, the code continuously reads accelerometer data from the X, Y, and Z axes, transmitted as a 6-byte buffer from the peripheral. The accelerometer values are parsed, processed, and sent via the **Serial Monitor**, which is used by the Edge Impulse platform to acquire the data necessary for training the Machine Learning model. The data is sent with the accelerometer value for each axis followed by a comma (",") because this is the standard format for data transmission via serial in Edge Impulse.
334
337
335
-
After establishing a connection, the code continuously reads accelerometer data from the X, Y, and Z axes, transmitted as a 6-byte buffer from the peripheral. The accelerometer values are parsed, processed, and sent via the Serial monitor, which is used by the Edge Impulse platform to acquire the data necessary for training the Machine Learning model. The data is sent with the accelerometer value for each axis followed by a comma (",") because this is the standard format for data transmission via serial in Edge Impulse. The protocol is straightforward: the device should send data at a baud rate of 115,200, with one line per reading, and individual sensor data separated by either a comma (",") or a TAB. If the connection is lost, the system handles the disconnection and restarts the scanning process to find another compatible device.
338
+
The **protocol** is straightforward: the device should send data at a baud rate of 115,200, with one line per reading, and individual sensor data separated by either a comma (",") or a TAB. If the connection is lost, the system handles the disconnection and restarts the scanning process to find another compatible device.
336
339
337
340
```arduino
338
341
void loop() {
@@ -408,14 +411,16 @@ void loop() {
408
411
delay(500);
409
412
}
410
413
```
411
-
With both codes running — the Nicla Sense ME code functioning as a peripheral, sending accelerometer data via BLE, and the Opta™ code acting as a central device, receiving this data via BLE and transmitting the readings to the computer via serial communication over a USB connection — you need a computer with internet access to forward the data to the Edge Impulse platform. To do this, run the command `edge-impulse-data-forwarder` in the terminal. Once the program is running and the necessary configurations are selected, the computer will forward the data to the Edge Impulse platform. You can then access your project under the `Data Acquisition` tab to start collecting data samples. For more information, refer to the official [Edge Impulse documentation](https://docs.edgeimpulse.com/docs/tools/edge-impulse-cli/cli-data-forwarder).
414
+
With both codes running, the Nicla Sense ME code functioning as a peripheral, sending accelerometer data via BLE, and the Opta™ code acting as a central device, receiving this data via BLE and transmitting the readings to the computer via serial communication over a USB connection, a computer with internet access is needed to forward the data to the Edge Impulse platform.
412
415
413
-
### Opta™ Deployment Code
416
+
To do this, run the command `edge-impulse-data-forwarder` in the terminal. Once the program is running and the necessary configurations are selected, the computer will forward the data to the Edge Impulse platform. You can then access your project under the `Data Acquisition` tab to start collecting data samples. For more information, refer to the official [Edge Impulse documentation](https://docs.edgeimpulse.com/docs/tools/edge-impulse-cli/cli-data-forwarder).
To deploy a trained Edge Impulse model, go to the `Deployment` tab in the Edge Impulse project, select `Arduino Library`, and click `Build` to generate the complete library. Download the generated library `Your_project_name.zip` file and import it into Arduino IDE by selecting `Sketch` > `Include Library` > Add `Your_project_name.ZIP` Library. This allows the model to run locally on the device without an internet connection, reducing latency, enhancing real-time performance, and minimizing power consumption. For more details, visit the [Edge Impulse documentation](https://docs.edgeimpulse.com/docs/run-inference/arduino-library#download-the-arduino-library).
After the deployment, it was possible to create the code that integrates Bluetooth Low Energy (BLE) functionality with Edge Impulse's machine learning inferencing framework to classify accelerometer data in real-time. It starts by initializing BLE communication and scanning for a specific peripheral device broadcasting a service with a known UUID. Once the device is found, it connects, discovers its attributes, and accesses a characteristic that provides accelerometer data. The code continuously reads 6 bytes of accelerometer data (representing the X, Y, and Z axes) from the BLE characteristic, processes it, and stores it in a buffer. This data is then transformed into a signal format compatible with Edge Impulse's inferencing engine. Using the pre-trained machine learning model deployed as a custom library generated by Edge Impulse's deployment feature, the code classifies accelerometer data, specifically vibration data, to detect anomalies. If the BLE connection is lost, the program automatically attempts to reconnect. Throughout the process, debug information and classification results are printed to the Serial monitor for real-time feedback.
420
425
421
426
The code is divided into four parts for better understanding, similar to what was done previously.
0 commit comments