Test Results

Objective: Acquire sensor response data from two different brands of hand sanitizer using a standardized testing procedure to train our algorithm. Determine if the hand sanitizers contain different concentrations of Acetone, Ethanol, and Methane detectable by the sensors and if residual VOCs affect further readings.

Setup: Figaro VOC sensors utilize a variable voltage divider circuit to translate chemical concentration into measurable voltage differences. When in the presence of a chemical, a variable resistor will change the output voltage accordingly. For this test, three sensors were placed in a mason jar with hand sanitizer for one minute and output voltage values were recorded. After one minute, sensors were removed from the mason jar and voltage values were once again recorded. This process was repeated for 10 minutes for two different brands of hand sanitizer. Voltage values were measured using Arduino Uno analog pins, at 10 second intervals. Sensors were allowed to rest for 20 minutes between testing. Testing occurred outside to allow airflow.

Once we obtained our STM32 MCU’s, we remade our code within the ST IDE to utilize the three available Analog to Digital Converters (ADCs) and we refined our testing method to the following:

• Warm up sensors for 5 minutes

• Collect ambient air VOC data for 1 minute

• Isolate sensor circuit with hand sanitizer and collect data for 5 minutes

• Remove hand sanitizer and collect ambient air data for 1 minute

• Repeat procedure with second type of hand sanitizer after 2 minutes

• Repeat procedure with ambient air for a set of reference VOC data

Results: Isolating each sensor in a glass mason jar containing one of two brands of hand sanitizers, Germ-X or Equate, resulted in different voltage readings. Each brand of hand sanitizer had a unique yet similar maximum output sensor voltage that can be used to train our machine learning algorithm. For the Figaro 2602 sensor, Germ-X hand sanitizer had an average maximum output voltage of 3.824V, and exposure to Equate hand sanitizer had an average maximum output voltage of 4.466V. When the Figaro 2612 sensor was exposed to Germ-X hand sanitizer, it resulted in an average maximum output voltage reading of 2.72V, and when it was exposed to Equate hand sanitizer, it resulted in an average maximum output voltage of 2.8V. The Figaro 2620 sensor had an average maximum output voltage of 3.901V when exposed to Germ-X, and an average maximum output voltage of 4.29V when exposed to Equate hand sanitizer.

Conclusion: The hand sanitizers contain different concentrations of Acetone, Ethanol, and Methane that are detectable by the sensors. We will be using this standardized testing method to acquire VOC data and will continue to improve the procedure as needed.

Objective: Connect the touchscreen to a programming interface and implement simple to use GUIs with limited functionality to see how usable they are when interacted with.

Setup: Using the Arduino IDE a simple GUI was created using Python. An Arduino was connected to my laptop and connected to the LCD display via jumper wires, the display had pins soldered into its 8-bit mode and was connected to a breadboard.

Results: Using Python we were able to easily interface with the touchscreen and create a simple preliminary layout. This is primarily because the libraries for the screen exist directly from Adafruit for use on Arduino. We were able to add functionality to the buttons and even configured it to light an external LED connected to the Arduino

Conclusion: This test is still in progress but we found that creating the GUI and adding the functionality is just a matter of creating different shapes at different coordinates of the touchscreen and enabling touches in those regions to correspond to a function. The next step is to recreate this for the STM32. The issue we foresee is finding libraries compatible with the STIDE.

Objective: Create a simple but functional multi-page GUI capable of: initiating and displaying the collection of sensor data, managing the external memory card, and allowing users to set up timers to help with their testing schedule process. The GUI should occupy as little space as possible, to save memory and reduce what’s being communicated to the LCD over SPI. It is navigated via a Dpad.

Setup: The LCD was set up and connected via 6 pins from the MCU. It communicates through SPI in its Full-Duplex Master mode where data is transmitted at a rate of 40 Mbps. Dpad buttons were created by creating GPIO inputs via GPIO_EXTI. Using the ST IDE and C, multiple pages were created through the use of a library capable of drawing shapes and displaying text through our LCD. Functionality is achieved by creating and programming functions to correspond with certain actions, and programming different visual effects on the screen when said actions occur.

Results: The LCD is capable of displaying all the pages I created in the C program and is fully navigable using a Dpad. The user can see where they are via a button animation, a separate animation is played when a selection is made. In total the GUI occupies about 59.3KB of memory, half of which is due to the library needed.

Conclusion: The GUI is fully functional and successfully serves its development purposes, any more additions needed should be easy to implement.

Objective: Design a housing to enclose all of the components in our project. The case needs to be compact, be structurally sound, and feature a way to easily access our VOC sensors. Design custom printed circuit boards (PCBs) to manage connections between peripherals and the microcontroller.

Setup: A preliminary case model was made in Fusion 360 and fabricated using Dr. Shrestha’s 3D printer. Once all necessary hardware components for the project were assembled, a revision of the preliminary design was created and fabricated using Dr. Shrestha’s 3D printer. Custom PCBs were designed in Eagle and fabricated using an online vendor.

Results: We have designed a case capable of housing all of the main components of our project, is easy to hold, and was designed with ways to easily replace VOC sensors, charge the battery and access different ports.The case dimensions were within our requirements, and the snap fit joints fit tightly together. We were also able to design and create custom PCBs to hold several subsystems together.

Conclusion: Our case is comparable in size to typical handheld devices, and enables us to easily access and replace various parts. We can easily update the design to accommodate new hardware components.

Objective: Upload the machine learning algorithm onto the microcontroller and integrate it with the other subsystems so that it is able to take real time sensor data as inputs and output a classification on the GUI.

Setup: We trained the ML algorithm using sensor data acquired, until we had reached our target accuracy level of 75%. Then using ST’s Cube AI software package, we imported and validated the algorithm. After the algorithm had been uploaded, we modified the IDE generated function MX_X_CUBE_AI_Process, to take the sensor data array as its input, run the algorithm, generate an output, and then display the results on our GUI. The function is called at the end of each sensor data collection interval so the classification is dynamic and is updated each time sensor data is acquired.

Results: The algorithm has been successfully implemented. Results are displayed on the GUI, and continuously update each time new sensor data is acquired. We were able to integrate a major component of our project with the other subsystems to form a cohesive system. When used in a non-controlled testing environment, such as without the mason jars, the output is still reliable. Running a test in open air showed that our algorithm was easily able to determine the classification. When a sample of Germ-X was introduced, the air classification percentage went down and the Germ-X classifications began to rise. The Germ-X classification percentage was significantly larger than the other two classifications.

Conclusion: By going through the process of creating and implementing a machine learning algorithm, we have demonstrated that our device can be used for classifying VOCs. The ease of which new algorithms can be implemented now that the framework has been created shows that our device could one day be used to run more advanced algorithms.

Objective: Measure the dimensions and weight of the constructed device to verify it’s of a size comparable to existing solution and is indeed portable.

Setup: A ruler was used to measure the device's dimensions, and a scale (Fig. 62)to measure its weight. The device, complete with all components, was placed on a scale and weighed. The resulting weight will be advertised as the weight of the device. A similar process was used for determining the device’s dimensions.

Results: The device’s dimensions and weight is within our specified parameters, and overall similar to that of a typical handheld device. The weight was 373g and had dimensions of 7 inches long, 5 inches wide, and 2 inches tall.

Conclusion: Our device is comparable in size and weight to a typical handheld device and within our engineering requirements.