"The industrial Internet of things (IOT) is brewing a broad transformation, which will not only make the mutual detection between interconnected machines a competitive advantage, but also make it an essential basic service. The industrial Internet of things takes the edge node as the starting point, and the latter is the target entry point of detection and measurement. This is where the physical world interacts with computational data analysis. Interconnected industrial machines can detect a large amount of information for key decision-making. This edge sensor may be far away from the cloud server for storage history analysis. It must be connected through a gateway that aggregates edge data to the Internet. Ideally, edge sensor nodes have a small size and can be easily deployed in space constrained environments.
Detection, measurement, interpretation, connection
In the first part of this multi-part industrial Internet of things series, we will decompose and study the basic aspects of edge node detection and measurement capability in the large-scale Internet of things framework: detecting, measuring, interpreting and connecting data, while also considering power management and security. Each section will explore a unique set of challenges. The intelligent management of edge nodes is very important for successful implementation. In some cases, ultra low power consumption (ULP) is the most important performance index. During critical events, when the sensor wakes up from sleep mode, most of the potential data can be filtered out.
Figure 1. Edge node devices intelligently detect, measure and interpret data and connect it to an internet gateway connected to the cloud. Data can be preprocessed through some forms of analysis, and then transmitted for deeper data mining and intelligent analysis.
Sensors constitute the front edge of the electronic ecosystem of the industrial Internet of things. The measurement phase converts the detected information into meaningful data, such as quantifiable values of pressure, displacement or rotation. In the interpretation stage, edge analysis and processing will convert the measured data into operable events. 1 only the most valuable information should be connected to the cloud across nodes for prediction or historical processing. In the whole signal chain, the data can be suppressed or filtered according to the initial acceptability limit. Ideally, sensor nodes should only send absolutely necessary information, and should make key decisions as soon as possible after obtaining key data.
Edge nodes must be connected to the external network through wired or wireless sensor nodes (WSNs). In this part of the signal chain, data integrity remains critical. If the communication is inconsistent, lost or damaged, there is little value in optimizing the detection and measurement data. Data loss during communication is not acceptable. The industrial environment with electrical noise can be very harsh and difficult, especially when RF communication is carried out in the presence of a large number of metal objects. Therefore, robust communication protocols must be designed in advance during system architecture design.
The power management of ULP system starts with the selection of regulator elements to achieve maximum efficiency. However, since the edge node may also wake up and sleep with a fast duty cycle, the power on and power off time should also be considered. External triggers or wake-up commands can help quickly remind edge nodes to start detecting and measuring data.
Data security is also an issue that must be considered in industrial Internet of things system. We not only need to ensure that the data within the edge is secure, but also ensure that its access to the network gateway is free from malicious attacks. Counterfeiting edge nodes to obtain network access for illegal activities is never allowed.
Intelligence begins at the edge
There are many detection solutions at the edge, which may not be just a single discrete device. There may be many different unrelated data acquisition at the edge. Temperature, sound, vibration, pressure, humidity, motion, pollutants, audio and video are only some of the variables that can be detected. These data will be processed and sent to the cloud through the gateway for further historical and predictive analysis.
It is no exaggeration to say that sensors are the backbone of the industrial Internet of things. But more accurately, they are the central nervous system for gaining insight. Edge node detection and measurement technology is the "birthplace" of target data. If bad or wrong data is truthfully recorded at this stage of the solution chain, no amount of post-processing in the cloud can recover the lost value.
Mission critical systems, such as healthcare and plant shutdown monitoring systems with high-risk results, require robust integrity of quality data measurements. Data quality is critical. False positives or omissions can be costly, time-consuming, and even life-threatening. Costly mistakes will eventually lead to unplanned maintenance, inefficient use of labor, and even have to disable the whole Internet of things system. Intelligence begins at the edge node, and the old saying applies here: if the input is garbage, the output must be garbage.
Figure 2. Many wired and wireless edge node outputs can be autonomously connected to the gateway node for aggregation before transmission to the ECS.
Being able to access the data treasure also means taking great responsibility
In the traditional signal chain solution without edge node intelligence, data is only data. Non intelligent nodes will never help generate wisdom and knowledge for making feasible decisions. 1 there may be a large amount of raw low-quality data that has no impact on the target performance of the system. 3 converting all this data and sending it to the final cloud storage destination may consume a lot of power and bandwidth.
In contrast, smart intelligent partition edge node detection and measurement will convert data into actionable information. Intelligent nodes can reduce overall power consumption, shorten latency and reduce bandwidth waste. This enables the reactive Internet of things with long delay to be transformed into a real-time predictive Internet of things mode. The Internet of things still applies the basic analog signal chain circuit design concept. For complex systems, it is usually necessary to have deep application expertise to interpret the processed data.
Optimized intelligent zoning maximizes cloud value
Only the most important measurement information needs to be sent to the cloud through the gateway for final processing. In some cases, most data is not important at all. However, for the time critical system data required for local real-time decision-making, we should act according to it as soon as possible before aggregating it to the remote node that can be accessed remotely. On the contrary, using historical values to influence long-term insight through prediction models is an ideal application of cloud processing. By archiving data to a huge database for retrospective processing and decision-making, it gives full play to the powerful advantages of cloud processing and storage. six
Figure 3. Intelligent partitioning of edge nodes solves new challenges that could not be solved before. The earlier simplified processing and intelligence in the signal chain realize a more efficient overall Internet of things solution.
Real time decision depends on the edge
IOT sensors are mainly analog sensors. The specific industrial application requirements will determine the dynamic range and bandwidth of the sensor required by the front end of the edge node. Before the signal is converted into digital representation and transmitted to the outside of the edge, the front end of the signal chain will be in the analog domain. If not selected properly, each element in the analog signal chain may limit the overall performance of the edge node. The dynamic range will be the difference between the target full-scale sensor and the background noise or the next highest useless signal.
Since IOT sensors often look for both known and unknown activities, analog filters are not always meaningful. Digital filtering is performed after sampling the signal. Unless an analog filter is used at the front end of the sensor, the harmonic or other spurious signals of the fundamental wave may mix with the detected information and compete with the target signal for power. Therefore, in the design stage, a response plan should be formulated for the unexpected detection signals in time domain and frequency domain to prevent interference artifacts from appearing in the measurement data.
The detected information is usually measured by the next ADC in the signal chain. If discrete components are used to design the edge nodes of the Internet of things, care should be taken not to reduce the dynamic range of the sensor when selecting the measurement ADC. The input full-scale range of the embedded ADC usually matches the output amplitude of the sensor well. Ideally, the sensor output should consume almost the entire ADC input range (within 1 dB) without saturating the ADC and being trimmed at the range limit. However, the amplifier stage can also be used to gain or attenuate the sensor output signal in order to maximize the dynamic range of the ADC itself. ADC full-scale input, sampling rate, bit resolution, input bandwidth and noise density will affect the signal measurement performance of edge nodes.
The front-end amplifier can be embedded in the measurement stage of the node or placed in front of the ADC as a discrete element. The gain, bandwidth and noise of the amplifier can also enhance the performance of the edge node.
The measurement ADC after the sensor in the signal chain usually adopts one of the following two sampling architecture types: Nyquist rate or continuous time - (CTSD), which is more common in embedded ADC. The Nyquist rate ADC has a flat nominal noise bottom equal to half the sampling rate frequency (i.e. FS / 2). CTSD combines oversampling rate and notch passband to make the noise exceed the target bandwidth, so as to increase the dynamic range. When understanding the analog bandwidth and dynamic range of edge nodes, it is very important to measure the ADC architecture and its resolution.
Figure 4. If there is no front-end analog filter on the IOT sensor, the Nyquist rate ADC will fold the higher-order frequency beyond the first Nyquist region back into the target bandwidth. In contrast, the CTSD ADC architecture with oversampling modulation clock uses noise shaping to achieve high dynamic range in the target frequency band. CTSD is not sensitive to signal aliasing because of its inherent filtering ability.
For example, in the frequency domain, the noise density of 1 Hz unit bandwidth will be based on the SNR of the ADC and the distribution width of the noise on the ADC sampling spectrum. In Nyquist rate ADC, the noise spectral density (per 1 Hz bandwidth) is 0 dB – ADC signal-to-noise ratio (SNR) – 10 × Log (FS / 2), where FS / 2 is the sampling rate divided by two or the single Nyquist region of the ADC. The ideal SNR is calculated as SNR = 6.02 × N + 1.76 dB, where n is the number of ADC bits. However, the actual SNR of ADC involves non ideal factors of transistor and semiconductor processing, including electrical noise and transistor level component defects. These non ideal factors will cause SNR performance to drop below ideal performance, so please refer to ADC data manual for SNR target performance.
The dynamic range of the edge node will consist of the dynamic range of the sensor, the amplification of the signal (if required) and the full-scale dynamic range of the ADC. If the full-scale output signal of the sensor does not reach within 1 dB of the input of the ADC full-scale range, part of the dynamic range of the ADC will be idle. Conversely, if the input from the sensor exceeds the range of the ADC, the sampled signal will be distorted. When calculating the dynamic range of edge nodes, amplifier bandwidth, gain and noise also need to be considered. The total electrical noise of the sensor, amplifier and ADC will be the square root of the sum of squares of each RMS component. seven
Figure 5. Example of loss of dynamic range (blue) due to mismatch between sensor signal output amplitude and ADC input full scale. Amplifiers are required to maximize the dynamic range of the sensor while preventing ADC saturation (red). Signal matching must consider the bandwidth, dynamic range and noise of the whole edge node signal chain.
Smart factory
In the industrial Internet of things, machine vibration monitoring will be a very important application. New or traditional machinery and equipment may have multiple key mechanical components, such as rotating shafts or gears, which may be equipped with MEMS accelerometers with high dynamic range. 8 these multi axis sensors will sample the vibration displacement of the machinery in real time. After measurement, the vibration signal can be processed and compared with the ideal machine configuration. 9 in the factory, the analysis of such information can help improve efficiency, reduce downtime and predict mechanical faults in advance. In extreme cases, the machine whose mechanical components are deteriorating rapidly can be shut down quickly to avoid further damage.
Figure 6. Although routine machine maintenance can be performed regularly, it is usually not performed intelligently according to the machine condition. 10 by analyzing the vibration performance of a specific machine operation, the predicted fault point and maintenance milestone warning can be issued at the edge node.
By implementing edge node analysis, the decision-making time delay can be significantly reduced. Figure 7 shows an example in which the system sends a warning immediately after exceeding the MEMS sensor warning threshold limit. If the event is extremely serious and is identified as a critical event, the node can be authorized to automatically disable the offending equipment to prevent a very time-consuming catastrophic mechanical failure.
Fig. 7. Time domain representation of machine vibration sampling data, where the comparator threshold determines whether to transmit detection and measurement data outside the edge. The system can maintain a low-power state to filter most of the information until data advantage is achieved through threshold cross events.
Alternatively, a trigger signal can be invoked to enable another detection and measurement node (such as a node on a standby machine element) to start interpreting the data according to the first event. This can reduce the total amount of sampling data from edge nodes. To determine any vibration anomaly relative to the nominal value, the front-end node must be designed to achieve the required detection performance. The dynamic range, sampling rate and input bandwidth of the detection and measurement circuit shall be sufficient to identify any offset events.
smart city
Another industrial IOT edge node application is a smart city industrial camera with embedded video analysis. According to the definition of smart city, the mission of the city is to gather countless information and communication points into a cohesive system to realize the management of urban assets. A common application is to provide parking vacancy reminder and occupancy detection. During commissioning, the field of view will be determined in advance for each camera. In the analysis mechanism, boundary edge detection can be defined and used to identify various objects and their motions. At the edge, not only the historical motion of the object can be analyzed, but also the digital signal processing (DSP) algorithm can be used to calculate the predicted path according to the object trajectory.
Figure 8. Using edge node video analysis, object type detection, trajectory and boundary crossing can be determined in a low-power system without sending full bandwidth video data to the cloud for analysis. Just transfer the timestamp and trace object coordinates and types.
In similar frequency filtering, the terminal usually does not need full bandwidth video analysis frames. Generally, when not used for security purposes, only a small part of the complete video frame is required. On a fixed camera, most of the visual data between frames is static data. Static data can be filtered out. In some cases, you only need to analyze the number of boundary crossings or motion coordinates of the target object. The reduced information subset can be transmitted to the next gateway in the signal chain in the form of trace coordinates.
Edge node video analysis can provide a variety of filter interpretation to distinguish various object types, such as cars, trucks, bicycles, humans and animals. This extraction operation reduces the data bandwidth and computing power required on the ECS. If you want to analyze the full frame rate video data sent downstream, it will occupy a lot of data bandwidth and computing power.
The indoor camera application can identify the number of people crossing the entrance boundary and adjust the lighting, heating or cooling of the room. To achieve visual effectiveness under extreme lighting conditions or other challenging lighting conditions, such as rainfall, outdoor cameras may need to have a high dynamic range. Typical composition of 8 or 10 bits per pixel
Our other product:
