A bearing condition monitoring method and system based on self-powered signal adaptive control
By synchronously acquiring signals with external sensors through a self-generating unit, generating dual-source diagnostic features, and automatically matching the transmission level, the problems of sensor failure, energy waste, and low diagnostic reliability in self-powered bearing monitoring are solved, thus achieving efficient bearing condition monitoring.
Patent Information
- Application Number
- CN202611233249.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-14
- Publication Date
- 2026-11-10
AI Technical Summary
In existing self-powered bearing monitoring solutions, sensors are prone to failure, consume a lot of energy, occupy a lot of wireless bandwidth, and are improperly utilized, resulting in decreased reliability of diagnostic results and information loss. Furthermore, the host computer lacks energy coordination and cannot actively request high-resolution data or fill in missing segments.
The system generates electrical signals through a self-generating unit and collects them synchronously with external sensor signals to generate dual-source diagnostic features. Based on the energy output level and the degree of fault abnormality, the system automatically matches the transmission level, sends feature data packets, and integrates the diagnostic features in the host computer for comprehensive judgment and control.
It enables automatic calibration for sensor drift or failure, optimizes energy utilization, reduces the number of sensors, improves diagnostic reliability and system survivability, achieves optimal balance between energy and information, and avoids missed diagnoses and waste.
Smart Images

Figure CN123065757A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of bearing condition monitoring technology, and in particular relates to a bearing condition monitoring method and system based on self-powered signal adaptive control. Background Technology
[0002] In long-term online monitoring scenarios for bearings in industrial rotating equipment, to overcome the drawbacks of external cables and battery power, such as complex wiring, regular battery replacement and maintenance, and battery life degradation in high-temperature and oily environments, the industry is gradually adopting self-powered bearing monitoring solutions. These solutions rely on generators to collect the mechanical energy of the bearing's operation and convert it into electrical energy to drive the monitoring nodes continuously. The implementation logic of existing conventional self-powered bearing monitoring solutions is as follows: a generator structure is configured to power the nodes, and independent vibration and temperature sensors serve as the sole source of fault data. The nodes transmit sensor feature data packets according to a fixed preset period and a fixed data integrity level.
[0003] The existing technical solutions have the following technical problems: Problem 1: Once an independent sensor fails, the node immediately becomes a "dumb node." In industrial scenarios with high temperatures, strong vibrations, and long lifespans, accelerometers and temperature sensors are prone to zero-point drift, sensitivity decay, and even failure. Furthermore, the nodes are located within rotating components, making on-site calibration and replacement difficult, leading to a decrease in the reliability of diagnostic results. While known patents such as CN103208880A involve multi-sensor bearing monitoring, none provide a mechanism for cross-calibrating and fault compensation of independent sensors based on the generated signal itself.
[0004] Question 2: Numerous sensors consume valuable energy and wireless bandwidth. Simultaneously acquiring multi-dimensional parameters such as rotational speed, vibration amplitude, temperature, and load requires existing solutions to configure multiple independent sensors, including accelerometers, Hall effect sensors, strain gauges, and temperature sensors. Each sensor consumes energy from the self-generating unit and occupies wireless transmission bandwidth, further exacerbating the "energy-constrained—information loss" dilemma.
[0005] Question 3: The output power of the self-generating unit varies drastically with rotational speed and vibration amplitude, and may even be unable to generate power during start-up and shutdown. Most existing solutions use fixed sampling rates and upload cycles, which either cause frequent disconnections due to power consumption exceeding the energy budget under low-energy conditions, or waste available energy and miss early fault characteristics due to conservative strategies under high-energy conditions. Known patents CN106891090A, CN109067277B, and CN109687635A only involve the generating mechanism itself and do not address "dynamic adjustment of feature upload strategies based on real-time energy budgets."
[0006] Question 4: The host computer passively receives data, lacking energy coordination with the terminal side. Most existing host computer software only performs passive data reception, plotting, and threshold alarms, failing to form a closed loop with the terminal side's energy budget. When energy is sufficient, it cannot actively request high-resolution data; when energy is insufficient, it cannot reconstruct and complete missing segments based on historical models.
[0007] Therefore, a solution is needed that can address at least one of the above problems. Summary of the Invention
[0008] The purpose of this invention is to provide a bearing condition monitoring method and system based on self-powered signal adaptive control, so as to solve the problems mentioned in the background art.
[0009] To achieve the above objectives, one technical solution adopted by the present invention is: a bearing condition monitoring method based on self-powered signal adaptive control, comprising the following steps: S1: Synchronous acquisition of power generation signals and bearing sensing signals: At least two different self-generating units generate electrical signals following the bearing rotation, and the electrical signals output by all self-generating units are acquired synchronously; at the same time, the bearing status sensing signals output by external sensing components are acquired. S2: Generate independent diagnostic features: Perform signal analysis on the electrical signals generated by the multi-channel self-generating unit, reverse the bearing operating conditions and fault characteristics, and obtain the first set of diagnostic features; extract the state features from the bearing sensing signals collected by the external sensing components to obtain the second set of diagnostic features. S3: Based on the energy output level represented by the real-time output power and / or energy storage level of the self-generating unit, and the degree of fault abnormality reflected by the first set of diagnostic features and the second set of diagnostic features, automatically match the feature transmission level corresponding to the information integrity, and send the feature data packet under the corresponding level to the outside. Preferably, the following steps are also included: S4: Multi-source feature collaborative processing and feedback control: Receives feature data packets output from each gear, integrates the first group of diagnostic features and the second group of diagnostic features to complete the comprehensive judgment of bearing status; generates control commands based on the comprehensive judgment results and sends them back to the acquisition end to adjust the transmission gear.
[0010] Preferably, the energy output level is determined by integrating the instantaneous output power of the self-generating unit through a sliding window to obtain the average power, and / or by sampling the energy storage level in real time.
[0011] Preferably, before fusing the two sets of diagnostic features, the two sets of diagnostic features are subjected to dimensionality normalization, and the feature deviation is calculated to characterize the degree of fault abnormality.
[0012] Preferably, multiple transmission levels with different information completeness are pre-configured; when the transmission level corresponding to the feature data packet is not greater than a preset level threshold, the missing feature information in the data packet is completed and reconstructed based on the historical running dataset.
[0013] Preferably, when analyzing the electrical signal of the self-generated unit, two types of feature extraction logic are included: extracting periodic features through frequency domain transformation to deduce the bearing operating speed; and extracting the signal envelope amplitude to characterize the bearing vibration strength.
[0014] Preferably, before sending the feature data packets outward, the data packets are converted to a communication format to adapt to the data parsing standard of the existing bearing monitoring equipment.
[0015] This solution also provides a bearing condition monitoring system based on self-powered signal adaptive control, which includes a data acquisition terminal and a host processing terminal; The acquisition terminal is installed on the bearing or its matching rotating component, and includes at least two self-generating units for outputting electrical signals, as well as an external sensing component for acquiring bearing status sensing signals, an energy storage management unit for outputting energy storage levels, a main control unit, and a data transmission unit. The main control unit synchronously collects various electrical signals, generates a first set of diagnostic features and a second set of diagnostic features, and combines real-time output power, energy storage level, and fault abnormality degree to match the transmission level of the corresponding information integrity, and sends the feature data packet through the data transmission unit. The host processing terminal receives feature data packets for each gear position, merges the two sets of diagnostic features to determine the bearing operating status, and sends a gear adjustment command to the acquisition terminal. Preferably, at least one self-generating unit employs an electromagnetic induction power generation structure, and at least one self-generating unit employs a variable capacitor electrostatic induction power generation structure, wherein the variable capacitor electrostatic induction power generation structure is configured with a polymer film carrying a constant charge and paired electrodes.
[0016] Preferably, the system is further configured with a cloud data center, which is communicatively connected to the host processing terminal and is used to receive and store diagnostic conclusions and corresponding feature data uploaded by each device; the cloud data center aggregates model parameters based on data from multiple devices and sends the updated diagnostic model back to the host processing terminal or the acquisition terminal.
[0017] The beneficial effects of this invention are as follows: This solution generates an electrical signal through a self-generating unit to obtain a first set of diagnostic features, which, together with a second set of diagnostic features from the external sensing component, constitutes a dual-source independent feature set. By calculating the feature deviation between the two and setting continuous judgment conditions, the system can accurately identify whether the external sensing component has drifted or physically failed. Once the failure of the external sensing component is confirmed, the system can automatically switch the main diagnostic signal source to the first diagnostic feature derived from the self-generated signal for output. This effectively solves the engineering pain point in existing solutions where the node immediately becomes a "dumb node" once the external sensor is damaged, greatly improving the survival rate and reliability of long-term online monitoring in harsh industrial environments. Through continuous monitoring of the deviation, the first diagnostic feature not only serves as a backup main source but also as a "touchstone" for the health status of the external sensor, providing a direct verification basis for the host computer to determine the reliability of the data.
[0018] This solution automatically matches the characteristic transmission level corresponding to the information integrity based on the energy output level represented by the real-time output power and / or energy storage level of the self-generated unit, as well as the degree of dual-source fault anomaly. This effectively solves the contradiction faced by self-powered nodes: "large energy fluctuations leading to the disconnection of fixed upload strategies, or wasting information under high energy conditions." It maintains liveness (uploading compressed packages) during low energy periods and achieves high fidelity (uploading complete spectrum) during high energy and abnormal conditions, realizing an optimal balance between energy utilization and data utilization. The host computer no longer passively receives and unilaterally sends alarms, but instead receives characteristic data packets, fuses dual-source features to complete a comprehensive judgment, and actively generates control commands based on the judgment results to send back to the acquisition end to adjust the transmission level for the next stage. This closed-loop control enables the system to have "adaptive perception feedback" capabilities. For example, when a weak early fault is detected, the host computer can proactively request the node to increase the characteristic upload level to obtain high-resolution data, avoiding missed diagnoses; when energy is scarce, it can proactively send commands to "downgrade" to extend the system's survival time.
[0019] In this invention, the average power is obtained by integrating the instantaneous output power of the self-generating unit through a sliding window, and / or the energy output level is determined by real-time sampling of the energy storage level. This can accurately reflect the energy income and expenditure status at the current moment, provide a reliable data basis for the dynamic matching of transmission levels, and avoid misjudgment of transmission levels due to energy estimation deviations.
[0020] In this invention, the two sets of diagnostic features are normalized in dimension before being fused, and the feature deviation is calculated to characterize the degree of fault abnormality. This makes the features from different physical sources (power generation signals and sensor signals) comparable, and can accurately quantify the differences between the two sets of diagnostic features, thereby effectively identifying early weak fault signs and abnormal states such as sensor drift / failure.
[0021] By pre-configuring multiple transmission levels with different information completeness, and when the transmission level corresponding to the feature data packet is no greater than the preset level threshold, the missing feature information in the data packet is completed and reconstructed based on the historical operation dataset. This allows the upper-level processing end to obtain relatively complete spectrum information for diagnosis even when only a small amount of compressed feature data can be uploaded under low-energy conditions, which greatly improves the reliability of diagnosis under low-energy conditions.
[0022] In this invention, when analyzing the electrical signal of the self-generated unit, the periodic features are extracted through frequency domain transformation to deduce the bearing operating speed, and the bearing vibration intensity is characterized by extracting the signal envelope amplitude. This allows the two key bearing state parameters, speed and vibration, to be obtained without relying on independent speed and vibration sensors, reducing the number of independent sensors and thus saving valuable energy budget and wireless bandwidth resources. Attached Figure Description
[0023] Figure 1 This is a schematic diagram of the overall system architecture of the present invention; Figure 2 This is an example diagram of the assembly of the sensing end mechanical structure and bearing in this invention; Figure 3 This is a flowchart of the end-side signal acquisition and adaptive feature upload decision-making process in this invention; Figure 4 It demonstrates the state machine logic for gear shifting; Figure 5 This is the time-domain waveform of the electromagnetic coil output V_e(t); Figure 6 It is an FFT spectrum of an electromagnetic signal; Figure 7 It is the time-domain envelope waveform of the electret output V_d(t); Figure 8 It is the speed inversion calibration fitting curve; Figure 9 It is the vibration RMS inversion calibration fitting curve; Figure 10 This is a diagram of the host computer software functional architecture; Figure 11 It is a flowchart of the upper computer's multi-source fusion, mutual calibration diagnosis and gear closed-loop control execution process; Figure 12 This is a system timing diagram; Figure 13 This is a diagram illustrating the accuracy of the rotational speed inversion. Figure 14 This is a graph showing the residuals from the rotational speed inversion. Figure 15 This is a diagram illustrating the accuracy of vibration inversion; Figure 16This is a diagram of the vibration inversion residuals; Figure 17 This is a schematic diagram comparing the traditional and the present invention's RMS time series. Figure 18 This is a schematic diagram of kurtosis evolution; Figure 19 This is a schematic diagram illustrating the peak evolution of the outer ring fault characteristic frequency (BPFO). Figure 20 This is a schematic diagram illustrating the alarm triggered by dual-source difference in this invention; In the diagram: 1. Bearing outer ring; 2. Bearing inner ring; 3. Cage; 4. Permanent magnet array; 5. PCB; 6. Electret film; 7. Coil. Detailed Implementation
[0024] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention.
[0025] Summary of Implementation Examples: This invention provides a bearing condition monitoring method and system based on self-powered signal adaptive control. It is suitable for online health monitoring scenarios of high-value rotating machinery bearings such as wind turbine main bearings, electric vehicle drive motor bearings, machine tool spindles, and industrial robot reducers. The overall architecture consists of a three-level hardware system: sensing end (self-powered bearing sensing node), host computer / edge gateway, and cloud data center. It is equipped with five core software algorithm modules: dual-source signal inversion, energy budget adaptive transmission, multi-source fusion and mutual calibration, historical data reconstruction and completion, and cross-product protocol adaptation.
[0026] This invention provides three types of engineering implementation examples: one for high-speed operating conditions of electric vehicle drive motor bearings; another for low-speed, heavy-load operating conditions of wind turbine main bearings; and the third for precision monitoring of high-speed electric spindles in CNC machine tools. The following sections first provide a unified description of the general hardware and algorithm logic for all implementation examples, then provide hardware parameters, calibration data, and experimental verification processes for each example, while also explaining the definitions of custom terms one by one.
[0027] 1. The self-generating unit comprises two types of subdivided hardware: an electromagnetic induction power generation structure (electromagnetic power generation module) and a variable capacitor electrostatic induction power generation structure (electret power generation module). In this scheme, the electromagnetic module generates electricity by using a Halbach permanent magnet array and coils to cut magnetic lines of force. In this scheme, the electret module generates electricity by using a polymer film with a constant charge to change the electrode spacing to induce charge (selectable materials include CYTOP, FEP, PTFE, and PVDF) and paired electrodes to form a variable capacitor. The bearing vibration changes the distance between the film and the electrodes, and the external circuit induces charge flow, resulting in an output voltage V_d(t).
[0028] 2. Energy output level is characterized by two parameters: (1) the average power P_avg(t) obtained by integrating the instantaneous output power of the self-generating unit through a sliding window; (2) the energy storage capacitor level V_cap(t) and fault abnormality degree sampled in real time by the energy storage management unit. The two parameters are used to determine the remaining available energy of the node.
[0029] 3. Transmission levels: This application presents three adaptive upload modes (including S1 ultra-low power, S2 standard, and S3 full spectrum, which will be mentioned later). Different levels correspond to different information integrity, data packet length, and operating power consumption. The level switching is jointly driven by the energy output level and the degree of fault anomaly.
[0030] 4. First group of diagnostic features: operating conditions and fault features obtained by inverting the output waveform of the self-generating unit, including speed, vibration RMS, misalignment coefficient, load torque, and bearing fault characteristic frequency (BPFO / BPFI / BSF / FTF).
[0031] 5. Second set of diagnostic features: raw sensing features acquired by external sensing components, including triaxial vibration acceleration time-domain / frequency-domain indices and bearing temperature time-series data.
[0032] 6. Fault anomaly degree: σ (anomaly score) is the dual-source feature deviation, calculated by the difference between the L2 norms of the two sets of feature vectors after normalization. The higher the value, the greater the risk of bearing failure or the greater the probability of sensor drift failure.
[0033] 7. Feature data packets: Standardized data frames encapsulated under different transmission levels, which are converted into a communication format compatible with existing wired bearing monitoring equipment after protocol adaptation before uploading.
[0034] 8. Multi-source feature collaborative processing: The entire process of normalization and fusion, mutual calibration, and reconstruction of missing data for the first and second groups of diagnostic features by the host computer.
[0035] 9. Energy Storage Management Unit: Requires integration of MPPT (Maximum Power Point Tracking) circuit, rectifier bridge, energy storage capacitor, and LDO (Low Voltage Regulator) circuit. It is responsible for collecting power from two sources, regulating and storing energy, and outputting the real-time energy storage level V_cap.
[0036] 10. Cloud Data Center: Includes historical operating condition database, bearing digital twin model, and multi-device federated aggregation module, to complete full-domain data storage, diagnostic model iteration, and updated model distribution.
[0037] Among them, see Figure 1The bearing condition monitoring system based on self-powered signal adaptive regulation includes: a sensing end (self-powered bearing sensing node): hardware integrating two types of self-powered units, independent external sensing components, energy storage management unit, low-power MCU, wireless communication module, and protocol adaptation sub-module; transmitting feature data packets to the host computer via wireless link and receiving gear control commands issued by the host computer; Host computer / edge gateway: The host computer hardware adopts an industrial edge gateway industrial control computer, and the software has a four-layer functional architecture as follows: Figure 10 The upper-level computer software functional architecture diagram is shown below: The bottom layer, the communication and data layer, includes a protocol parsing unit, a time-series InfluxDB database, an object waveform storage unit, and a cloud synchronization API interface; the middle layer, the algorithm service layer, includes a signal completion and reconstruction unit, a dual-source mutual calibration engine, a working condition profile dictionary, and a CNN / Transformer diagnostic model library; the middle layer, the analysis and decision layer, includes working condition identification, early fault warning, remaining life (RUL) prediction, and transmission gear strategy distribution units; the top layer, the user interface layer, includes a large visualization screen, tiered alarm notifications, and historical data playback functions. The upper-level computer can directly interface with the applicant's existing wired vibration monitoring equipment diagnostic software. Only the protocol adaptation module needs to be enabled to reuse the existing mature algorithm libraries for envelope demodulation, order tracking, and automatic threshold alarms, without needing to reconstruct the diagnostic logic. Simultaneously, the upper-level computer reserves OPC-UA, Modbus, CANopen, and PROFINET industrial bus interfaces, which can issue equipment control commands to downstream frequency converters and lubrication controllers. Cloud Data Center: The cloud data center deploys a cloud server cluster, wirelessly connected to multiple field host computers. Internally, it is divided into three main functional units: a historical database storing the time-series characteristics of all equipment, fault tags, and 30-day continuous evolution test data; a digital twin model unit storing the speed, load, and temperature calibration mapping relationships of various bearing models; and a multi-device federated aggregation module collecting monitoring data from all nodes, completing the aggregation and iteration of fault diagnosis model parameters, and transmitting the updated model back to the field host computer or the local flash memory storage of the sensing MCU via the network. The overall signal flow of the three-layer hardware system is as follows: Figure 1 As shown in the overall system architecture diagram, the sensing end uploads feature data packets to the host computer via a BLE / LoRa / Sub-GHz wireless link. The host computer then transmits gear adjustment commands back to the sensing end via a policy distribution channel. Simultaneously, the host computer uploads complete diagnostic data to the cloud to complete model iteration. Application layer: Adapted to all types of rotary bearing monitoring scenarios, including wind turbine spindles, electric vehicle motor bearings, machine tool spindles, rail transit axle boxes, and industrial robot reducers.
[0038] Among them, see Figure 2The sensing end mechanical structure is fully integrated inside the standard rolling bearing body. The detailed assembly steps are as follows: the bearing balls are evenly distributed inside the bearing through the cage. A Halbach permanent magnet array is arranged on the outer ring of the bearing, and a coil is arranged on the inner ring of the bearing (i.e., forming an electromagnetic induction self-generating unit). When the inner and outer rings rotate relative to each other, the coil moves to cut the magnetic field lines and generates current. This embodiment provides an installation layout scheme, specifically: on the stepped surfaces on both sides of the raceway on the inner wall of the outer ring, a ring-shaped stepped groove is machined on each side corresponding to the size of the magnet. The permanent magnet array can be a "one-piece ring magnetic ring" or "multi-segment arc-shaped magnetic blocks" arranged according to Halbach polarity for seamless splicing. After the permanent magnet is embedded in the groove, it is fully bonded and fixed using high-strength epoxy resin structural adhesive. In order to resist the huge centrifugal force and axial movement when the bearing is running at high speed, a C-shaped retaining ring or retaining ring is also installed on the outer end face of the groove (the side away from the raceway) to firmly press the magnet into the groove. The coil is mounted on the stepped surfaces (shoulders) on both sides of the raceway on the outer wall of the inner ring of the bearing, maintaining a radial air gap that is coaxial and perpendicular to the permanent magnet array of the outer ring. The coil is made of enameled wire wound on a ring frame made of non-magnetic material (such as FR4 epoxy board or PPS plastic) to form a ring coil array with an iron core (or without an iron core). The whole is vacuum-encapsulated with epoxy resin to form a solid whole. Several axial lead wire through holes (wiring holes) are pre-drilled at the bottom of the stepped groove of the inner ring. The lead wires of the coil pass directly through these holes to the end face of the inner ring and are directly soldered to the ring PCB. The PCB is mounted on an axial end face of the bearing's inner ring. This is characterized by several threaded holes and a locating stop machined on the inner ring's end face. The PCB is designed as a ring with an inner diameter larger than the shaft's outer diameter. It fits directly onto the shaft, tightly against the inner ring's end face, and is secured to the inner ring's threaded holes by 3-6 high-strength, durable studs (or countersunk screws). To ensure dynamic balance during high-speed rotation, the screws must be symmetrically distributed and dynamically balanced. If space is limited, the electret film can be directly integrated onto the bottom surface of the ring-shaped PCB. That is, the bottom layer of the PCB is a comb-shaped electrode, followed by an electret polymer film (such as CYTOP or FEP), a counter electrode layer, and finally encapsulated by an extremely thin insulating protective layer. Alternatively, an insulating substrate can be pre-attached to the innermost layer of the inner ring (near the step in the bearing's inner bore), and then the electret film can be placed on top and fixed with a miniature pressure plate. Because the film is extremely thin (micrometer-level thickness), it occupies almost no space inside the bearing; all components, including the triaxial MEMS accelerometer, temperature sensor, MCU, energy storage capacitor, and wireless communication chip, are surface-mount soldered onto the surface of the ring-shaped PCB; the energy storage capacitor is a low-ESR supercapacitor with a capacitance range of 100μF~1mF; the PCB has reserved outgoing contact points for connecting the MPPT energy storage management circuit and the rectifier bridge.
[0039] The following section will introduce the end-side signal acquisition, energy budget calculation, and adaptive transmission mode execution process in this scheme (i.e., the bearing condition monitoring method based on self-powered signal adaptive control), as detailed below: (See reference...) Figure 3 The end-side signal acquisition and adaptive feature upload decision-making flowchart shows that the MCU synchronously and in parallel acquires two power generation signals and independent sensor signals, performs energy budget estimation and feature extraction in parallel, and finally switches the transmission level through the adaptive decision-making module. The complete execution steps are as follows: The first step is to synchronously acquire signals: the MCU synchronously samples the electromagnetic coil output V_e(t) and the electret thin film output V_d(t) through dual channels, and synchronously acquires the triaxial accelerations a_x, a_y, a_z and the bearing temperature T(t); the two generation signals are connected to the rectifier MPPT circuit of the energy storage management unit to acquire the instantaneous generation voltages V_e and V_d, currents I_e and I_d in real time, and simultaneously sample the node load voltage V_load, load current I_load and energy storage capacitor level V_cap.
[0040] The second step involves calculating the real-time energy budget P_avg(t), corresponding to the integral formula defined in the claim: P_avg(t) = (1 / T) ×∫[tT, t] (V_e·I_e + V_d·I_d - V_load·I_load)dt, where T is the duration of the sliding integration window, which is uniformly taken as 1s in this embodiment; V_e·I_e is the instantaneous output power of electromagnetic power generation; V_d·I_d is the instantaneous output power of electret power generation; V_load·I_load is the instantaneous power consumption of the total load of the sensing circuit; the integration result divided by the window duration yields the average available energy budget P_avg(t), which is used as one of the core judgment parameters for gear switching.
[0041] The third step involves parallel extraction of two types of features: One type is the second group of diagnostic feature extraction, which performs time-domain statistics, FFT frequency-domain transformation, and envelope demodulation on the raw triaxial acceleration and temperature signals to extract sensor feature vectors F_A such as vibration kurtosis, temperature mean, and spectral peak value. One category involves the extraction of diagnostic features for the first group. Frequency domain transformation and envelope amplitude extraction are performed on V_e(t) and V_d(t) respectively to complete the operating condition inversion. The complete inversion calibration logic and measured data are explained in modules below: Figure 5 This is the time-domain waveform of the electromagnetic coil output V_e(t). The test speed is 1500 rpm. The horizontal axis is time in ms, and the vertical axis is output voltage V. The waveform is a standard periodic sine wave, which is the original input signal for speed inversion.
[0042] Figure 6It is an FFT spectrum of electromagnetic signal, with frequency (Hz) on the horizontal axis and amplitude on the vertical axis. The main frequency f_e=300.0Hz is extracted as the core parameter for speed inversion, corresponding to the measured speed of 1500rpm. Figure 7 It is the time-domain envelope waveform of the electret output V_d(t), which is the defect modulation waveform of the bearing outer ring BPFO=89.2Hz. The envelope amplitude directly characterizes the intensity of vibration. Figure 8 This is the speed inversion calibration fitting curve. The horizontal axis is the electromagnetic main frequency f_e, and the vertical axis is the measured rpm of the encoder. An FFT transformation is performed on the output voltage V_e(t) of the electromagnetic power generation module to extract the main frequency f_e. Therefore, the speed rpm = c1·f_e + c0. In the embodiment using a 12-pole Halbach array, the calibration constants are c1 = 5.0053, c0 = -1.8442, and the goodness of fit R0 is... 2 =0.99999, the horizontal axis 0~600Hz corresponds to the speed 0~3200rpm, the scatter points are the measured calibration data, the dashed line is the fitting curve, and the maximum inversion error is less than 5rpm; Figure 9 This is the vibration RMS inversion calibration fitting curve, with the horizontal axis representing the electret envelope amplitude V_d,env, and the vertical axis representing the measured effective vibration value (m / s) from the excitation table. 2 The envelope of the electret output V_d(t) is extracted, and its envelope amplitude V_d,env has a linear mapping with the bearing vibration RMS, denoted as RMS_vib = k·V_d,env + b. The measured calibration constant k of this invention is 85.11 (m·s). -2 ) / V, b = -1.0756 m·s -2 R 2 = 0.9979, 5m / s 2 The inversion error within the measurement range is less than 5%.
[0043] The calibration relationship completes the calculation of all the first group of diagnostic features: the main frequency f_e is extracted by V_e(t) FFT and substituted into the speed formula to calculate the real-time speed; the ratio of the second harmonic of V_e(t) to the main frequency amplitude R_2 is extracted to determine the misalignment fault. In this scheme, a critical value is set to determine the misalignment fault when R_2 is greater than a critical value, such as 0.3, 0.35, 0.4, etc. The value selection is the critical value determined by the batch fault test system, which is reflected in the following steps: Step 1, multiple groups of normal coaxial working condition bench tests, batch calculation of R_2 for each group, and statistically the maximum value of R_2 under normal conditions is about a; Step 2, gradually apply coupling offset and shaft eccentricity to simulate misalignment faults, and continuously record the changes of R_2; Step 3, when the eccentricity reaches the industrial allowable fault level, R_2 stably exceeds the value b; Step 4, comprehensively consider noise, instantaneous impact interference margin, and finally determine the engineering judgment threshold b.
[0044] The fourth step is to extract the envelope amplitude V_d(t) and substitute it into the formula to obtain the effective value of vibration RMS; demodulate the envelope spectrum of V_e(t) to extract the fault characteristic frequencies of BPFO, BPFI, BSF, and FTF; map the instantaneous current amplitude of the rectifier circuit to the bearing load torque, and integrate all parameters to form the power generation inversion characteristic vector F_B.
[0045] The fifth step is to calculate the anomaly score σ = ‖F_A-F_B‖. Before the calculation, the two sets of feature vectors F_A and F_B are normalized to eliminate the differences in the numerical values of different dimensions of acceleration, voltage and rotation speed, and to quantify the degree of deviation between the two sets of signals.
[0046] Step 6: Adaptive transmission gear determination; the gear switching state machine logic is as follows: Figure 4 As shown in the energy budget adaptive state machine, the pseudocode for state machine transitions is as follows: if V_cap < V_min: state = S1 # Energy storage voltage is below the minimum safety threshold, urgently switch to ultra-low power mode; elif σ>= σ_th and P_avg >= P2: state = S3 # Fault abnormality exceeds the standard and energy is sufficient, switch to full spectrum high-fidelity mode; elif P_avg >= P1 and V_cap >= V1: state = S2 # Energy meets the standard and there is no obvious fault, run standard transmission mode; else: state = S1 # Energy is insufficient, maintain ultra-low power mode.
[0047] Typical threshold parameters for the unified state machine in this embodiment are: P1=0.5mW, P2=5mW, V1=2.4V, V_=1.8V, σ_th=0.30. All thresholds support dynamic modification via remote issuance from the host computer. The differentiated transmission content, data packet length, and measured power consumption for the three modes are shown in Table 1 below. Table 1: model Triggering conditions Uploaded content Typical package length Actual average power consumption S1 Ultra-Low Power P_avg < P1 or V_cap < V1 4 time-domain indicators + 16-byte compressed signature Approximately 16 bytes 4.8 µW S2 Standard Mode P_avg ≥ P1 and V_cap ≥ V1 Time-domain indicators + Top-10 envelope spectra + inversion features Approximately 64 bytes 38.6 µW S3 Full Spectrum Mode P_avg ≥ P2 and σ ≥ σ_th Complete spectrum + original waveform clip 1~4 KB 285.4 µW Step 7: After completing the above process, the feature data packet is encapsulated and transmitted: The MCU packages the corresponding feature data according to the selected level, and sends it to the protocol adaptation submodule before transmission to complete the communication format conversion. The wireless data frame is converted into the Modbus RTU / TCP format compatible with existing wired monitoring devices, and then uploaded to the host computer edge gateway through the BLE / LoRa / Sub-GHz wireless module.
[0048] See Figure 11 Next, we will introduce the execution process of multi-source fusion, mutual calibration diagnosis, and gear closed-loop control of the host computer: Step 1: Data packet parsing and timing alignment, distinguishing between S1, S2 and S3 data packets, and extracting timestamps, packet identifiers and original feature codes; Step 2: After the host computer receives the feature data packet uploaded by the sensing end, it first reads the preset gear identifier field in the header of the data packet to distinguish the transmission mode. If the data packet is identified as S2 standard mode or S3 full spectrum mode, it directly skips the spectrum reconstruction and completion related calculation process, and directly parses the complete time domain indicators, envelope peak sequence and even the original full spectrum waveform in the data packet. It extracts the second set of diagnostic feature vectors F_A corresponding to the external sensing component and the first set of diagnostic feature vectors F_B obtained by inverting the self-generated signal. After extraction, it is directly sent to the multi-source fusion mutual calibration engine to carry out subsequent anomaly scoring calculation and fault judgment. There is no need to generate an approximate spectrum to supplement the missing information. The S2 mode data packet fully carries the time domain indicators, the first 10 peaks of the envelope spectrum and all power generation inversion features. The core frequency domain information required for fault identification is not lost. The S3 mode directly transmits the uncompressed original complete spectrum waveform. The data dimension and frequency point information are complete and without loss. Skipping the reconstruction process can also reduce the computing power occupation of the host computer and shorten the overall diagnostic response latency.
[0049] If the data packet header identifier indicates S1 ultra-low power mode, the host computer will automatically initiate a compressed feature spectrum reconstruction and completion pipeline based on a conditional generative adversarial network (cGAN) or variational autoencoder (VAE). This pipeline performs a complete completion operation to compensate for the significant loss of spectral information under low-energy conditions. Before formally executing model inference operations, the host computer first performs S1 data packet splitting and preprocessing, separating four basic time-domain indicators and a 16-byte binary compressed feature code (c) from the data packet. Simultaneously, it retrieves the real-time operating parameters corresponding to this data transmission synchronization, including bearing real-time speed, load rate, operating temperature, energy storage capacitor voltage, and bearing model. Operating parameters with significant differences in dimensions and numerical ranges are standardized and normalized, and concatenated to generate a one-dimensional conditional vector (w) required for model computation. The 16-byte binary compressed feature code is then decoded and restored, converting the quantized and compressed spectral peak positions and amplitude range identifiers into low-dimensional digital latent vectors recognizable by the model. After completing all input data preprocessing, the host computer retrieves the offline-trained model weight file pre-stored on the local hard drive for a single preprocessing step. For inference operations, in the cGAN architecture implementation, the generator network receives the preprocessed condition vector w and the compressed feature decoding latent vector in parallel via two paths. After dimensionality upscaling and channel fusion, the two features are output through multi-layer deconvolution operations, resulting in a 1024-point complete reconstructed approximate spectrum X1 whose dimensions, frequency intervals, and sampling intervals are completely aligned with the actual spectrum acquired in S3 mode. During the training phase, the cGAN network is iterated collaboratively by the generator and discriminator, using the condition vector as a global constraint, and alternately updating the two sets of network weights. At the same time, adversarial loss and L1 reconstruction loss of spectral amplitude are superimposed to control the fitting accuracy until the average reconstruction error of the test set samples is stably controlled within 8%, at which point all weights are frozen. In the VAE architecture implementation, the encoder maps the spliced condition vector and compressed feature latent vector to a Gaussian distribution latent space. The decoder synchronously accesses the sampled latent space variables and the same set of condition vectors to reconstruct and output the complete reconstructed spectrum X1. During the training phase, iterative convergence is achieved by relying on KL divergence regularization loss and L2 reconstruction loss of the spectrum. Similarly, the reconstruction error of the test set ≤8% is used as the criterion for saving the model.
[0050] After the model inference generates the reconstructed spectrum X1, the host computer retrieves the historical real reference spectrum X stored in the local time series database under the same operating conditions. ref ,according to The average relative error is calculated using a frequency-by-frequency error check to verify the usability of the reconstructed spectrum. For example, the calculated average relative error... A distortion rate of no more than 8% indicates that the distortion of the reconstructed spectrum meets the accuracy requirements for industrial fault diagnosis. The host computer then uses this reconstructed spectrum to replace the complete measured spectrum for a full set of feature extraction operations. From the reconstructed spectrum, the envelope spectrum sequence, four types of bearing fault characteristic frequencies (BPFO, BPFI, BSF, and FTF), vibration RMS effective value, and shaft misalignment amplitude ratio R_2 are extracted sequentially. These are integrated to obtain the first set of diagnostic features F_B. Simultaneously, the second set of sensing features F_A is generated by combining the four time-domain indicators from the S1 data package. The two sets of normalized feature vectors are simultaneously sent to the mutual calibration engine to calculate the anomaly score σ and complete the bearing health status determination. The 8% error upper limit threshold comes from the statistical results of batch samples under full working conditions on the test bench. During the offline training phase, a total of 100 sets of triplet samples (w,c,X) covering the entire speed, load, and fault status are collected. 70% of the samples are used for model training, and 30% are used for... The sample is used as a test set. If the maximum average reconstruction error calculated from all test samples is only 7.92%, and an 8% margin is reserved, it is determined as the reliable judgment boundary. At the same time, this error threshold can ensure that the accuracy of fault characteristic frequency point amplitude identification is maintained above 98.5%, avoiding missed or false alarms due to spectrum distortion of early weak faults.
[0051] After each reconstruction operation is completed, the host computer will also synchronously store the corresponding working condition vector, 16-byte compressed feature code, and generated reconstruction spectrum into the local historical working condition database to continuously expand the working condition sample library to support subsequent incremental optimization and iteration of the model.
[0052] After feature vector extraction or spectrum reconstruction and completion, the system enters the dual-source mutual calibration operation stage. Before the operation, the two feature vectors F_A and F_B are normalized to eliminate the difference in dimensions. Then, the anomaly score σ = ||F_A-F_B||2 is obtained by calculating the vector L2 norm. To avoid misjudgment in a single sampling caused by instantaneous impact and electromagnetic noise, the system continuously collects the anomaly scores of 5 sampling windows for statistical calculation to obtain the mean value μ of σ. σ With standard deviation S σ The selection of five consecutive windows for statistical smoothing is an engineering parameter determined by the combined bearing vibration signal fluctuation period and the wireless transmission jitter period. The duration of a single window matches the basic sampling period of the node. The statistical analysis of five consecutive samples can filter out random interference signals. Subsequently, according to (σ-μ) σ S σThe criterion of ≥3 times the standard deviation is used to determine that the current anomaly is a real and valid equipment anomaly. 3 times the standard deviation is a classic statistical low-probability criterion. Under normal operating conditions, the probability of the deviation caused by random noise exceeding this threshold is less than 0.3%, which can greatly reduce the probability of false alarms. After determining the valid anomaly, the system executes differentiated control logic in three scenarios. Among them, 0.30 is the basic fault judgment threshold for the anomaly score σ. This value comes from the statistics of bench tests on multi-gradient shaft misalignment and bearing pitting failure. Under fault-free and stable operating conditions, the dual-source characteristic deviation is generally lower than 0.30. When a fault occurs or the sensor drifts, σ will stably exceed this value, which serves as the stratified judgment benchmark. The first scenario involves a short-duration, pulse-like increase in σ, with only a few sampling windows exceeding the limit. This indicates an early, weak fault in the bearing. The host computer's internal strategy distribution module generates a gear-switching control command, which is transmitted back to the sensing end's main control MCU via a wireless communication link. This forces the device to switch to S3 full-spectrum mode to acquire high-resolution, complete waveforms, accurately capturing early fault characteristics. The second scenario involves σ consistently exceeding 0.30 for multiple consecutive sampling windows. This indicates zero-point drift or physical damage to the independent triaxial accelerometer and temperature sensor. The host computer automatically masks the second set of features F_A from the sensors and uses only the first set of features F_B obtained from the generator signal inversion as the sole primary diagnostic source to output the monitoring results. This addresses the deficiency of existing solutions where nodes become "dumb nodes" after sensor failure. The third scenario involves σ consistently falling below 0.30 for an extended period. This indicates that the external sensors are functioning normally and the bearing has no obvious potential faults. The system integrates the two sets of feature vectors, F_A and F_B, to generate a comprehensive diagnostic conclusion, improving the reliability of diagnostic results under normal operating conditions.
[0053] After completing dual-source mutual calibration and fault determination, the process proceeds to output diagnostic conclusions and synchronize with the cloud. The host computer integrates all diagnostic information, including the current bearing health status, remaining RUL service life, and fault type labels. Through the built-in protocol adaptation module, the data is encapsulated into the required compatible data format and synchronously pushed to the original diagnostic visualization interface and hierarchical alarm channel. At the same time, the host computer uploads all raw feature data and the final diagnostic results to the cloud data center in real time. The cloud platform performs federated aggregation calculations based on the full-domain monitoring data uploaded by multiple field devices, iteratively optimizing the internal parameters of the fault diagnosis model. After optimization, the updated model is sent to the field host computer or directly stored in the local flash memory of the sensing MCU, realizing continuous iterative optimization of the full-domain device diagnostic capabilities.
[0054] See Figure 8The system's timing diagram fully reproduces the entire process of adaptive energy control combined with automatic fault diagnosis: The bearing starts running and generates vibration, which drives two types of self-generating modules to output electrical energy. After the voltage of the energy storage capacitor continuously rises to the V_on wake-up threshold, it triggers the MCU to start working. The MCU simultaneously performs signal acquisition and real-time energy budget assessment. After determining that the current energy level meets the operating conditions of the S2 standard mode, it uploads a standard feature data packet through the wireless module. The host computer receives the data packet, completes the parsing and dual-source feature comparison, calculates the abnormal score σ exceeding the standard, and then sends a gear switching command to the lower-end sensing end to force a switch to the S3 full-spectrum mode. After receiving the control command, the MCU collects the complete spectrum waveform and uploads it to the host computer again. The host computer completes the fault diagnosis based on the complete waveform and simultaneously uploads all monitoring data to the cloud. After the cloud completes the update of the diagnostic model parameters, it sends the data back to the field sensing node, forming a closed-loop adaptive control system.
[0055] The following will explain the details of the calibration tests and fault evolution measurements for three types of application scenarios. Example 1: High-speed monitoring scenario of 6206 motor bearing, as detailed below: The servo motor torque test bench was used to complete the 11-speed calibration test. The complete speed inversion calibration data is shown in Table 2 below (corresponding to...). Figure 13 , Figure 14 ): Table 2: Target setting rpm actual encoder RPM test Electromagnetic frequency fe (Hz) Formula inversion rpm Inversion error rpm 100 98.97 19.393 95.22 -3.75 200 198.54 40.246 199.6 1.06 400 397.78 80.677 401.97 4.19 600 600.73 120.17 599.65 -1.09 900 899.68 180.049 899.36 -0.32 1200 1199.97 240.068 1199.78 -0.19 1500 1499.88 300.49 1502.21 2.33 1800 1801.7 359.979 1799.97 -1.72 2200 2200.39 440.099 2201 0.61 2700 2702.63 539.729 2699.68 -2.95 Ten vibration amplitude calibrations were performed using a vibration table. The complete vibration RMS inversion data is shown in Table 3 below (corresponding to...). Figure 15 , Figure 16 ): Table 3: Actual RMS (m / s2) measured on the vibration table Electret envelope Vd,env(V) Formula inversion to calculate RMS (m / s²) Inversion error m / s2 0.5 0.0185 0.499 -0.001 1 0.0242 0.981 -0.019 2 0.039 2.246 0.246 3.5 0.0503 3.202 -0.298 5 0.0712 4.982 -0.018 7.5 0.0959 7.09 -0.41 10 0.1398 10.827 0.827 15 0.1922 15.284 0.284 20 0.2385 19.221 -0.779 30 0.3671 30.169 0.169 The steady-state temperature rise calibration data under different loads (ambient temperature 24.0℃, 1500rpm) are shown in Table 4 below: Table 4: Load rate % Bearing steady-state temperature (°C) Temperature rise difference K 10 33.22 9.22 25 35.93 11.93 40 38.94 14.94 55 41.53 17.53 70 44.7 20.7 85 46.94 22.94 100 49.25 25.25 The applicant also conducted a fault evolution test, which showed the following results: The outer ring of the 6206 bearing was pre-introduced with an electrical discharge machining (EDM) to create a 0.4×0.6mm micro-pitting defect. After 30 days of continuous operation at 1500rpm and 0.6kW load, data was collected every 6 hours. The 30-day evolution tracking results are as follows: Figures 17-20 The traditional ISO10816 vibration RMS threshold is 4.5 m / s². 2An alarm was triggered on the 26th day. The dual-source anomaly score σ of this invention exceeded the threshold of 0.30 for the first time on the 14th day, providing an early fault warning 11.8 days in advance. After the alarm, the host computer issued a command to switch to the S3 full-spectrum mode. The envelope spectrum clearly identified the BPFO=89.21Hz fundamental frequency and multiple harmonics. The test piece was disassembled to verify the accuracy of the outer ring defect diagnosis conclusion.
[0056] It should be further explained that the Halbach permanent magnet array, ring coil, electret film and ring PCB integrated structure built into the inner and outer rings of the bearing in this invention are equipped with a complete dynamic balancing correction process in the design, processing and assembly stages. It will not introduce additional eccentricity, additional vibration or new fatigue defects into the bearing. The whole structure adopts a symmetrical ring layout, with permanent magnet blocks, coils, circuit boards and thin film components evenly distributed along the circumference. The mass and radial thickness of the components at each position are completely consistent. After assembly, they are uniformly subjected to double-sided correction on a dynamic balancing machine, and the overall residual imbalance is eliminated by micro-weight matching.
[0057] Example 2: Low-speed heavy-load monitoring scenario for the main bearing of wind turbine SRB240 / 630: The bearing operates within a speed range of 8-22 rpm. The electromagnetic power generation module employs a variable reluctance topology to ensure low-speed power generation capability, achieving a stable average output of 1.2mW at 22 rpm. Wireless transmission utilizes a LoRa SX1262 module, with a single gateway covering multiple wind turbine sensor nodes within a 3km radius. The steady-state average power consumption of a single node is 22µW, enabling continuous operation without an external power supply. In a fault injection test, a 0.5×0.5mm pitting defect was detected on the inner ring. The dual-source mutual calibration identification alarm time of this solution is 19 days earlier than that of traditional wired sensing solutions. Under shaft misalignment conditions, the abnormal score σ increases significantly within 6 hours, instantly triggering a full-spectrum acquisition command from the host computer.
[0058] Example 3: High-speed electric spindle bearing monitoring scenario for CNC machine tools: The spindle has a maximum speed of 24,000 rpm, and the electromagnetic coil is miniaturized to an outer diameter of 8 mm. The MCU used is the low-power STM32L4R5. After the spindle is powered off during shutdown, it relies on the electret module to provide power to maintain short-term status monitoring. The power consumption at the 18,000 rpm high-speed steady-state condition node is 80 µW. For the weak FTF frequency fault of cage detachment, the traditional single acceleration sensor solution is prone to missed detection. This invention can reliably identify early cage abnormalities by inverting the FTF characteristics of the generated signal and combining dual-source deviation calibration.
[0059] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.
Claims
1. A bearing condition monitoring method based on self-powered signal adaptive control, characterized in that: Includes the following steps: S1: Synchronous acquisition of power generation signals and bearing sensing signals: At least two different self-generating units generate electrical signals following the bearing rotation, and the electrical signals output by all self-generating units are acquired synchronously; at the same time, the bearing status sensing signals output by external sensing components are acquired. S2: Generate independent diagnostic features: Perform signal analysis on the electrical signals generated by the multi-channel self-generating unit, reverse the bearing operating conditions and fault characteristics, and obtain the first set of diagnostic features; extract the state features from the bearing sensing signals collected by the external sensing components to obtain the second set of diagnostic features. S3: Calculate the feature deviation between the first diagnostic feature and the second diagnostic feature, and generate an anomaly score based on the feature deviation; when the anomaly score exceeds a preset threshold and meets the set conditions, automatically switch the main diagnostic information source of bearing condition diagnosis to the first diagnostic feature as the current monitoring result output.
2. The bearing condition monitoring method based on self-powered signal adaptive control according to claim 1, characterized in that: It also includes the following steps: S4: Based on the energy output level represented by the real-time output power and / or energy storage level of the self-generating unit, and the degree of fault abnormality reflected by the first set of diagnostic features and the second set of diagnostic features, automatically match the feature transmission level corresponding to the information integrity, and send the feature data packet under the corresponding level to the outside. S5: Multi-source feature collaborative processing and feedback control: Receives feature data packets output from each gear, integrates the first group of diagnostic features and the second group of diagnostic features to complete the comprehensive judgment of bearing status; generates control commands based on the comprehensive judgment results and sends them back to the acquisition end to adjust the transmission gear.
3. The bearing condition monitoring method based on self-powered signal adaptive control according to claim 2, characterized in that: The energy output level is determined by integrating the instantaneous output power of the self-generating unit through a sliding window to obtain the average power, and / or by sampling the energy storage level in real time.
4. The bearing condition monitoring method based on self-powered signal adaptive control according to claim 2, characterized in that: Before fusing the two sets of diagnostic features, the two sets of diagnostic features are normalized in dimension, and the feature deviation is calculated to characterize the degree of fault abnormality.
5. The bearing condition monitoring method based on self-powered signal adaptive control according to claim 2, characterized in that: Multiple transmission levels with different information completeness are pre-configured; when the transmission level corresponding to the feature data packet is not greater than a preset level threshold, the missing feature information in the data packet is completed and reconstructed based on the historical operation dataset.
6. The bearing condition monitoring method based on self-powered signal adaptive control according to claim 2, characterized in that: When analyzing the electrical signal of the self-generated unit, two types of feature extraction logic are involved: extracting periodic features through frequency domain transformation to deduce the bearing operating speed; and extracting the signal envelope amplitude to characterize the bearing vibration strength.
7. The bearing condition monitoring method based on self-powered signal adaptive control according to claim 2, characterized in that: Before sending feature data packets outward, the data packets are converted to a communication format to adapt to the data parsing standard of the existing bearing monitoring equipment.
8. A bearing condition monitoring system based on self-powered signal adaptive control, characterized in that: The system includes a data acquisition terminal and a host processing terminal; The acquisition terminal is installed on the bearing or its matching rotating component, and includes at least two self-generating units for outputting electrical signals, as well as an external sensing component for acquiring bearing status sensing signals, an energy storage management unit for outputting energy storage levels, a main control unit, and a data transmission unit. The main control unit synchronously collects various electrical signals, generates a first set of diagnostic features and a second set of diagnostic features, and combines real-time output power, energy storage level, and fault abnormality degree to match the transmission level of the corresponding information integrity, and sends the feature data packet through the data transmission unit. The host processing terminal receives feature data packets for each gear position, merges the two sets of diagnostic features to determine the bearing operating status, and sends a gear adjustment command to the acquisition terminal.
9. The bearing condition monitoring system based on self-powered signal adaptive control according to claim 8, characterized in that: At least one self-generating unit employs an electromagnetic induction power generation structure, and at least one self-generating unit employs a variable capacitor electrostatic induction power generation structure, wherein the variable capacitor electrostatic induction power generation structure is configured with a polymer film carrying a constant charge and paired electrodes.
10. The bearing condition monitoring system based on self-powered signal adaptive control according to claim 8, characterized in that: The system is also equipped with a cloud data center, which is communicatively connected to the host processing terminal and is used to receive and store diagnostic conclusions and corresponding feature data uploaded by each device. The cloud data center aggregates model parameters based on data from multiple devices and sends the updated diagnostic model back to the host processing terminal or the acquisition terminal.
Citation Information
Patent Citations
Magnetic suspension flywheel energy storage device and fault diagnosis method thereof
CN103208880A
Energy storage vibration mechanism
CN106891090A
Optimal field weakening control method based on the safety of the spring-loaded electric rudder system
CN109067277B
Nested-rotor open-core flywheel
CN109687635A
