1 Characteristics of BME688
- Can we use the current PCB using the BME680 print and upgrade it with BME688 without any changes?
Yes, the BME688 is fully backward compatible and can be placed on every PCB, which has been designed for the BME680. You just have to update the API to the new version, which automatically detects the BME688 and does the right value calculation (necessary due to the extended ASIC range of the BME688).
- Do you need to force air into the BME688 when taking samples?
No, this just happens “by itself” due to the laws of nature, in particular due to diffusion. As soon as the gas composition around the BME688 changes, there is a concentration gradient, which leads to diffusion of gases into and out of the BME housing. In addition, thanks to the tiny dimensions (the lid hole is less than 1 mm away from the gas sensor chip inside) this happens within few seconds.
- As the data is stored on the sensor, how many measurements are possible?
The BME688 stores the scan profile as well a few data points in its buffer. The measured data has to be continuously read by the microcontroller (MCU) of the device to which the BME688 is connected anyway. In that configuration, the only limit for measurement data is the device storage. By running the BME688 library (called “BSEC”) on the MCU, the sensor data can directly be evaluated. Therefore, you do not have to store raw data but only the required output values.
- How much computing power is required to run the AI software? Can you give examples for compatible MCUs?
BME AI-Studio requires a desktop PC to analyze the data and derive the best algorithm. However, running a defined algorithm is not demanding at all anymore. For instance an ESP8266 or ESP32 can do everything in real-time.
- Can you please explain how the gas scanner works?
The gas sensor is measuring with different sensitivities during one gas scan. By doing so, it can generate a specific fingerprint for different gas mixtures. In addition, you can modify and optimize the scan profile with BME AI-Studio on your application.
- Does the BME688 have Flash or EEPROM?
No.
2 AI firmware and algorithm of the BME688
- As soon as the algorithm is trained with data from several sensors in order to recognize one component, can we export the parameters and use only one sensor to detect it?
Yes, the BME AI-Studio software tool can export the trained algorithm as a configuration string, which can be loaded into the BSEC 2.0 software on any device having one BME688. This allows the device to directly output the trained scan results of the BME688.
- Do you use the sensor to train its AI model?
Yes, the sensor data is being used for the AI model. The standard gas scan mode for VSC detection is being developed based on sensor data from a huge number of sensors and lab tests with different gases. And for other applications, BME AI-Studio software tool enables everyone to develop own use-cases based on BME688 sensor data, for instance by using the dev-kit with eight BME688 sensors.
One of the major benefits of the BME688 is that you can directly use sensor data measured in real-life applications. So far, the typical procedure for enabling new gas sensor-based applications is the following:
- Identify single lead gases as well as potential interfering gases with sophisticated gas analyzers.
Find or develop suitable gas sensors for these gases. - Do lab tests with lead gases against interference gases (usually in synthetic air, which cannot represent real life conditions).
- Test reproducibility and performance in the real-life application.
This procedure still makes sense in case of known target gases like for e.g. sulfur compounds as a marker for bad breath. However, it comes to its limits for other smells or more complex gas mixtures.
With BME688 and the BME AI-Studio software tool you can directly develop, test and optimize in your application. For sure, this can still be accompanied by lab tests and might be even mandatory in some applications. However, using real-life data for gas sensing algorithms can significantly improve the performance and even enable new use-cases.
- Which neural network topology is used in AI? Or is it just statistical analysis for pattern recognition?
The current version of the BME AI-Studio software tool uses a pre-defined Neural Net Architecture combined with one configurable optimizer for training (ADAM optimizer). Depending on market requirements, we plan to have the possibility of choosing other architectures in a future release.
3 Applications and use cases of the BME688
- Is it possible to detect the gases produced by burning of network cables and electronics?
We expect that this can be an interesting application for using the BME688.
Let’s give some background: If there are unusual states in electric circuits or cabinets, there are typically two reasons:
- Materials (e.g. isolator) get hot due to high currents / shorts,
- High voltages or shorts lead to flashovers / sparks.
In case 1, hot or even melting materials produce increased outgassing, typically many unburned hydrocarbons, which can be well detected by the BME688 (just as humans smell it). In case 2, flashovers generate ozone, which is well detected by a BME688 and has a completely different signature than other gases.
- Does BME688 detect combustible gases like CH4 & C3H8 and poisonous gases like CO?
Yes, nearly all hydrocarbons (CxHy) are being detected by the BME688 gas sensor as well as many other gases like for example CO. Combustible gases are typically classified into NMOG (“Non-Methane Organic Gases”) and methane. Methane (CH4) is an exception, since its decomposition requires special catalysts, so we do not expect high signals even for high concentrations of methane. However, in many applications, methane does not appear as a single gas but together with other gases (e.g. sulfur compounds), which are well detected by BME688. Therefore, it makes sense to test with the BME688 in your application.
- Next to gasses, what else can be measured with the sensor?
The BME688 has temperature, barometric pressure, air humidity and gas sensor elements inside. All sensor information can be used either as single values or combined in the AI software to recognize certain conditions or states. In the BME AI-Studio software tool, you can decide whether you just want to use the gas sensor data for your application or take as well pressure, temperature and humidity sensor data into account.
- Are profiles of pre-trained models of some gases available for implementation?
The standard profile is developed to detect VSCs. There are several other gas scan profiles available in the BME AI-Studio software tool and you can even configure your own profile. For sure, they have to be trained on an application.
- Are you planning to offer more pre-trained models than currently available?
Bosch Sensortec, as well as first customers are already developing models for other applications.
- Is the BME688 certified as a medical sensor?
No, usually this is done on device level, not on sensor component level. Bosch Sensortec qualifies products according to the standard requirements for consumer electronics (e.g. JEDEC). Generally, our customers are the experts for their specific application field and mission profile in which they use our sensor components. End devices anyway have to be certified, the advantage of certified components is usually low.
- Is it possible to detect drugs and explosions with your BME688?
We do not have experience in this field. However, you can test it with the BME688 development kit.
- How can I find out if the gases in my application can be measured with the BME688?
This is exactly one reason why we have developed the BME AI-Studio software tool. You can test directly in your application in real life conditions. There is no need to know which gases in which concentrations might be target gases or which other gases might be present as well. Just get started.
So far, one typically had to start lab tests with single gases in synthetic air for every single application. However, even if this works in the lab, it does not mean that it works as well in the field, because in real life many other gases are present as well.
With the BME AI-Studio, you can develop, test and optimize in your real application.
4 Tools for BME688
4.1 Software tools
- Do you have a code and examples showing the integration and use with your BSEC library?
Both is becoming available on our website https://www.bosch-sensortec.com/software-tools/software/BME688-software/ for download until end of March.
- Are over-the-air updates possible to detect more gasses and substances in future?
You can do that even today: The configuration of the BME688 is completely defined by software, not hard-coded in the ASIC. For instance, if you have developed a new application with BME AI-Studio, the result is a new configuration string in combination with a scan profile. Both can be easily transferred to each of your devices in the field over the air, since it is just a few kb of size. As soon as the BSEC software on your device is loaded with the new configuration, the sensor works with the new characteristic.
- Does the AI software only use the gas sensor or could I use the humidity sensor data as well for detecting events?
Of course, the AI software can use all sensor data from the 4-in-1 sensor BME688: gas, humidity, temperature and pressure signals. You can choose which one is reasonable for your application.
4.2 Hardware tools
- Why do we need 8 sensors on the evaluation board?
The BME688 development kit can be configured with the BME AI-Studio software tool. This allows to optimize performance, ODR and power consumption on specific application needs. By featuring eight BME688 sensors, the board allows you to test and gather data with more than one configuration at the same time. This significantly increases statistics and reduces development time as well.
- Where is the BME development kit X8 (BME688) available?
Both BME688 and the development kit with eight BME688 sensors will be available at all our distribution partners as of April 2021. We are going to link to all suppliers on our BME688 product website.
