A self-generating type fault diagnosis and early warning system

Abstract

The application discloses a self-power generation type fault diagnosis early warning system, which comprises a composite energy collecting device, a power management energy storage module, a multi-dimensional sensing device and an edge computing diagnosis unit; the composite energy collecting device converts high-temperature waste heat and mechanical vibration energy into electric energy; the power management energy storage module adopts a hybrid energy storage architecture of supercapacitors and lithium sub-batteries to adjust and store and supply power for the system; the multi-dimensional sensing device and the edge computing diagnosis unit are arranged integrally, the multi-dimensional sensing device collects the displacement and three-axis acceleration of equipment and adaptively adjusts a detection period; the edge computing diagnosis unit performs on-site analysis based on VMD variational mode decomposition, a deep time sequence prediction network and an SVM support vector machine algorithm, deduces an operation situation in a future set time period, identifies fault feature diagnosis fault probability, early warning level and fault type and the like and outputs early warning information. The application solves the problem that a traditional sensor depends on a battery and realizes zero-delay accurate diagnosis by using edge computing.

Description

Technical Field

[0001] This invention belongs to the field of industrial Internet of Things monitoring and fault diagnosis technology, and relates to a self-generating fault diagnosis and early warning system; specifically, it relates to a self-generating fault diagnosis and early warning system that utilizes environmental temperature difference energy and mechanical vibration energy to achieve energy self-sufficiency, and integrates embedded edge computing algorithms to perform real-time fault mode recognition and early warning of the tested object. Background Technology

[0002] With the digital transformation of asset-heavy industries such as metallurgy, petrochemicals, and energy, the assessment of operational continuity of core rotating machinery and pipeline nodes is becoming increasingly critical. These high-value assets operate under harsh conditions of extreme heat, high pressure, and severe alternating loads for extended periods, and their health directly impacts overall production capacity and safety. Therefore, it is imperative to rely on high-frequency situational awareness and early warning mechanisms to build a safety barrier for industrial system operations.

[0003] However, existing industrial online monitoring technologies have three main drawbacks: First, limited energy supply: wired solutions are difficult to deploy in high-altitude, explosion-proof, and confined spaces and are prone to introducing new fault points; wireless nodes rely on chemical batteries, requiring frequent manual battery replacements for high-frequency continuous sampling, and are more likely to create monitoring blind spots due to battery depletion in uninhabited areas such as offshore wind farms and hazardous chemical plants; Second, limited sensing dimensions: existing solutions mostly monitor only single parameters such as vibration amplitude, lacking fusion analysis of multi-dimensional information such as displacement and triaxial acceleration, making it difficult to identify complex faults, resulting in high false alarm and false negative rates; Third, significant diagnostic delays: mainstream systems adopt a "data pass-through" mechanism, with terminals only blindly collecting data. Uploading massive amounts of raw waveforms to the cloud not only consumes radio frequency bandwidth and increases power consumption, but also makes it impossible to distinguish between operating condition fluctuations and actual faults locally in real time due to network latency. It can only output coarse over-limit alarms and is unable to provide specific fault probabilities, warning levels, fault types, and fault levels. Summary of the Invention

[0004] Purpose of the invention: The purpose of this invention is to provide a self-generating fault diagnosis and early warning system that can resolve the multiple dilemmas of existing industrial measurement and control systems, such as heavy reliance on external power supply, cumbersome on-site deployment, time-consuming manual inspection, and serious lag in fault identification.

[0005] The technical solution of the present invention is: a self-generating fault diagnosis and early warning system, comprising a composite energy harvesting device, a power management energy storage module, a multi-dimensional sensing device, an edge computing diagnostic unit, and a remote monitoring platform. The composite energy harvesting device is used to convert the thermal energy and mechanical vibration energy generated by the operation of the monitored equipment into electrical energy. The power management and energy storage module is used to regulate and store the electrical energy converted by the composite energy harvesting device, and to provide working power for the multi-dimensional sensing device and the edge computing diagnostic unit. The multi-dimensional sensing device is used to collect the displacement changes and triaxial acceleration changes of the monitored equipment in real time. It is integrated with the edge computing diagnostic unit and connected to the power management energy storage module through shielded wires. The edge computing diagnostic unit has a built-in high-performance, low-power microprocessor that runs a fault diagnosis algorithm. This unit is used to analyze the data collected by the multi-dimensional sensing and detection device on-site, identify the fault characteristics of the monitored equipment, including component loosening, wear and structural cracks, pressure instability and liquid erosion, diagnose the fault probability, warning level, fault type and fault level, predict the operating status of the monitored equipment in the future within a set time period, generate warning information, and send it to the remote monitoring terminal through the wireless communication module.

[0006] Furthermore, the composite energy harvesting device includes, from bottom to top, a thermoelectric power generation module, a multifunctional composite substrate, and an electromagnetic vibration power generation module. The cold end of the thermoelectric power generation module shares the same double-sided copper-clad aluminum nitride ceramic substrate with the multifunctional composite substrate. The copper layer on the lower surface of the double-sided copper-clad aluminum nitride ceramic substrate is etched with copper conductive sheets for welding P-type semiconductors and N-type semiconductors. A ceramic substrate is installed on the bottom side of the copper conductive sheets of the P-type semiconductors and N-type semiconductors. The copper layer on the upper surface of the double-sided copper-clad aluminum nitride ceramic substrate serves as the lower electrode layer of the multifunctional composite substrate. A piezoelectric unit and the upper electrode layer are coaxially bonded on top of it. An upper conductive ceramic substrate is insulated and bonded on top of the upper electrode layer. A metal spring fixing seat is rigidly installed at the center position above the upper conductive ceramic substrate. The electromagnetic vibration power generation module includes an elastic element and a permanent magnet. The top of the elastic element is equipped with a permanent magnet, and the bottom is rigidly connected to a metal spring fixing seat. An induction coil is installed around the permanent magnet and the elastic element. The induction coil installed around the permanent magnet and the elastic element, together with the piezoelectric unit on the multifunctional composite substrate, serve as a supplementary energy source. The multifunctional composite substrate has a double-sided functional structure, with its lower surface being a heat exchange surface and its upper surface being an electromechanical conversion surface; a magnetic base is also installed at the bottom of the multifunctional composite substrate. A magnetic shielding protective shell is installed around the thermoelectric power generation module, the multifunctional composite substrate and the electromagnetic vibration power generation module. A quick-release slot is provided on the side wall of the magnetic shielding protective shell, and a heat dissipation plate fin is installed in the quick-release slot.

[0007] Furthermore, the magnetic base is made of high-temperature resistant rare-earth permanent magnet material, and its contact surface is designed as an arc-shaped structure that adapts to the curvature of the surface of the monitored device. The double-sided copper-clad aluminum nitride ceramic substrate has both thermal conductivity and electrical insulation properties. The piezoelectric unit uses a ring-shaped or disc-shaped lead zirconate titanate piezoelectric ceramic sheet, which is coaxially attached to the upper surface of the lower electrode layer by conductive silver paste or low-temperature sintering process.

[0008] Furthermore, the material of the lead zirconate titanate piezoelectric ceramic sheet is selected according to the surface temperature conditions of the monitored device: When the surface temperature of the equipment is below 120℃, PZT-5H material with a high voltage coefficient should be selected. When the surface temperature of the device is higher than 120°C, PZT-4 or PZT-8 modified materials with high Curie temperature are selected, and the multifunctional composite substrate serves as the cold end of the thermoelectric power generation module, with its operating temperature lower than the depolarization temperature of the piezoelectric material.

[0009] Furthermore, the power management energy storage module includes a power management energy storage module housing, a power management energy storage module circuit board encapsulated inside the power management energy storage module housing, a lithium-ion battery, and a supercapacitor. The supercapacitor is integrated on the power management energy storage module circuit board, the lithium-ion battery is fixed to one side of the power management energy storage module circuit board and electrically connected to the circuit board through wires, and the power management energy storage module circuit board leads out to the external interface through the aviation socket of the power management energy storage module shell via shielded wires. The power management energy storage module adopts a multi-channel impedance matching input circuit and a hybrid energy storage architecture of supercapacitor and lithium-ion battery. The power management energy storage module is designed with independent rectification and buck-boost circuits for thermoelectric power generation (low voltage, high current), electromagnetic power generation (medium voltage, medium impedance), and piezoelectric power generation (high voltage, high impedance). The supercapacitor is used to buffer the pulse energy generated by vibration power generation and to provide instantaneous current for the wireless communication module to transmit data; The lithium-ion battery serves as a backup power source to maintain the system's low-power sleep mode and timed heartbeat function when the monitored equipment is shut down for an extended period without temperature differences or vibrations.

[0010] Furthermore, the multi-dimensional sensing device includes an internally installed 4G transmission module, displacement sensor, tilt sensor, and high-performance, low-power microprocessor. A multi-dimensional sensing device housing is installed outside the 4G transmission module, displacement sensor, tilt sensor and high-performance low-power microprocessor. An aviation connector and an antenna are respectively installed on the outer wall of the multi-dimensional sensing device housing. The multi-dimensional sensing device utilizes multi-source information fusion technology to cross-verify the displacement collected by the displacement sensor and the triaxial acceleration collected by the tilt sensor, thereby assisting the edge computing diagnostic unit in identifying complex faults.

[0011] Furthermore, the multi-dimensional sensing device supports bidirectional wireless communication, which is used to receive configuration instructions issued by the remote monitoring terminal to adjust the sampling frequency, alarm threshold or update the diagnostic model. The multidimensional sensing device also has the function of adaptively adjusting the detection cycle, including the following aspects: (1) Read the remaining voltage value and voltage change rate in the power management energy storage module to reflect the energy conversion efficiency of the composite energy harvesting device; (2) Using the data collected by the displacement sensor (11) and the tilt sensor (12), calculate the current vibration amplitude and vibration frequency of the monitored object to reflect the operating status of the monitored object; Based on the above-mentioned energy status parameters and equipment status parameters, a joint judgment is made. When the energy is sufficient and the equipment shows abnormal signs, the detection cycle is shortened and high-frequency diagnosis is switched. When the energy is sufficient and the equipment status is stable, the regular detection cycle is maintained. When the energy reserves are tight, the detection cycle is extended to prioritize ensuring the system is online. The housing of the multidimensional sensing device is made of high-strength engineering plastic or die-cast aluminum alloy, and its seams are equipped with sealing rings.

[0012] Furthermore, the displacement sensor, tilt sensor, and 4G transmission module in the multidimensional sensing device, along with the high-performance, low-power microprocessor in the edge computing diagnostic unit, are all arranged on the same PCB circuit board using a surface mount soldering method.

[0013] Furthermore, the edge computing diagnostic unit operates through a fault diagnosis and early warning algorithm, which is a combination of variational mode decomposition, deep temporal prediction network, and support vector machine, and includes the following steps: Step (1) Data preprocessing and slicing: The acquired raw triaxial acceleration and displacement signals are detrended, and the continuous signal is divided into several analysis samples using a sliding window mechanism. Overlap sampling is used to increase the sample density. Step (2) Adaptive signal decomposition: Using the variational mode decomposition algorithm, the number of modes K is adaptively determined by the center frequency observation method, and the non-stationary signal is decomposed into K bandwidth-limited intrinsic mode components, and the energy entropy of each IMF component is calculated; Step (3) Dynamic feature sequence construction: Extract the time domain statistical features and frequency domain features of each IMF component respectively, and fuse the displacement deviation values ​​collected by the displacement sensor. After Z-score standardization, construct a multi-dimensional feature vector and splice it in chronological order to form a time series feature matrix. Step (4) Multi-step prediction of future trends: Input the time series feature matrix into the built-in deep time series prediction network, set the span of multi-step prediction by modifying the HORIZON parameters, and deduce and output the prediction feature vector of the corresponding time node in the future. Step (5) Fault mode and probability assessment: Input the predicted feature vector into a support vector machine classifier based on particle swarm optimization; The classifier uses a radial basis function kernel to output the posterior probability of each fault type occurring in the future, i.e., the fault probability, and determines the fault level based on the magnitude of the deviation of the feature vector from the normal baseline. Step (6) Intelligent early warning fusion decision: Introduce DS evidence theory, map the output fault probability and the state of temperature and tilt sensors to the basic probability allocation function, and use Dempster synthesis rules to calculate the final fault confidence; combine fault confidence and fault level to generate alarm messages including fault probability, early warning level, fault type and fault level.

[0014] Furthermore, the operating method of the self-generating fault diagnosis and early warning system includes the following steps: Step (1) Multidimensional data acquisition: The displacement change and triaxial acceleration change of the monitored equipment are acquired in real time through a multidimensional sensing device; Step (2) Edge-side local diagnosis: The edge computing diagnosis unit reads the collected data and runs the fault diagnosis and early warning algorithm locally; firstly, it uses VMD variational mode decomposition to denoise and extract features from the original signal to construct a multi-dimensional feature vector; then, it inputs the feature vector into the built-in deep temporal prediction network in time sequence to deduce the predicted feature vector of the monitored device in the future set time period; then, it uses the built-in SVM support vector classifier to identify fault modes, and finally outputs the posterior probability of each fault type occurring in the future period. Step (3) Intelligent early warning decision: The edge computing diagnostic unit introduces the DS evidence theory, integrates the posterior probability output by SVM with the auxiliary sensing status, calculates the final fault confidence, and makes local decisions accordingly: if the device is determined to be operating normally, the wireless communication module is controlled to remain in sleep mode or only send heartbeat packets; if the device is determined to be faulty and the confidence exceeds the set threshold, early warning information containing fault probability, early warning level, fault type and fault level is immediately generated. Step (4) Remote wireless transmission: The edge computing diagnostic unit wakes up the wireless communication module and sends the warning information to the remote monitoring terminal; the remote monitoring terminal receives and displays the warning information for maintenance personnel to view.

[0015] This invention utilizes a longitudinally integrated layered structure of temperature difference, piezoelectricity, and electromagnetic to extract waste heat and mechanical vibration energy from equipment operation, achieving energy self-sufficiency and maintenance-free long-term operation. It integrates multi-dimensional sensors such as displacement and triaxial acceleration and employs multi-source information fusion technology to comprehensively capture the health status of the equipment. The fault diagnosis and early warning algorithm based on VMD + deep temporal prediction network + SVM is pre-executed at the edge and key diagnostic and early warning conclusions are uploaded, achieving zero-latency accurate early warning while significantly reducing communication power consumption and bandwidth usage.

[0016] Beneficial Effects: Compared with existing technologies, this invention has the following significant advantages: 1. Multi-source micro-energy power generation, achieving "passive and maintenance-free" operation throughout the entire life cycle: A multi-domain coupled energy trapping array of "thermal-mechanical-electric" is constructed, adopting a longitudinally integrated stacked structure of temperature difference-piezoelectric-electromagnetic. By efficiently absorbing waste heat and broadband vibrations from the surface of industrial equipment, energy self-sufficiency under all operating conditions is achieved, eliminating the labor costs and safety hazards caused by frequent battery replacements in traditional monitoring nodes, and realizing maintenance-free long-term operation of the sensor network; 2. Edge intelligent diagnosis, achieving zero-latency early warning: The fault diagnosis algorithm based on VMD + deep temporal prediction network + SVM is moved to the sensor end (edge ​​side), which can process massive amounts of raw data locally in real time; it not only significantly reduces wireless transmission power consumption, but also achieves immediate response and early warning for sudden faults, with security far exceeding that of traditional solutions that rely on cloud processing; 3. Targeted and simplified communication mechanism, overcoming the dual bottlenecks of "power consumption and bandwidth" in wireless transmission: By deploying edge computing diagnostic units, the interactive mode of sending massive, disordered raw high-frequency waveforms to the base station is replaced. Instead, data is only pushed to the cloud when a fault is diagnosed. High-value early warning messages and transient snapshots containing "fault probability, early warning level, fault type, and fault level"; it reduces wireless communication frequency and data throughput by more than 99%, alleviating RF power consumption pressure and network congestion problems in industrial sites; 4. Multimodal parameter cross-verification constructs extremely high fault diagnosis "confidence": abandoning the conventional single-parameter threshold monitoring mode, it introduces a multimodal sensing module that includes spatial displacement and tilt angle dynamic response; through deep cross-verification of multidimensional spatiotemporal data at the local algorithm layer, it effectively shields interference noise caused by non-stationary operating conditions, and can... It can more comprehensively and accurately assess the health status of core equipment and achieve a high degree of confidence in diagnosing complex faults; 5. Adaptive and non-destructive deployment, perfectly suited to complex and harsh heavy industry sites: Equipped with a high-strength magnetic anchor with curvature self-adaptation, it can quickly fit onto the surface of various irregular pipes or rotating housings without the need for shutdown or power outage, or destructive operations such as drilling and welding; Combined with an industrial-grade high-protection isolation chamber and strong and weak current zone shielding design, it not only has excellent resistance to strong vibration and detachment, but also can withstand the attack of oil, high temperature and electromagnetic surges for a long time, demonstrating excellent engineering adaptability. Attached Figure Description

[0017] Figure 1 This is the overall system framework diagram of the present invention; Figure 2 This is a cross-sectional view of the composite energy harvesting device in this invention; Figure 3 This is a schematic diagram of the overall structure of the composite energy harvesting device in this invention; Figure 4 This is a front view of the composite energy harvesting device in this invention; Figure 5 This is a schematic diagram of the structure of the multifunctional composite substrate in the composite energy harvesting device of the present invention; Figure 6 This is a schematic diagram of the electromagnetic vibration power generation module in the composite energy harvesting device of the present invention; Figure 7 This is a schematic diagram of the thermoelectric power generation module in the composite energy harvesting device of the present invention; Figure 8 This is a schematic diagram of the structure of the multidimensional sensing device in this invention; Figure 9 This is the internal circuit diagram of the multidimensional sensing device in this invention; Figure 10 This is a schematic diagram of the power management energy storage module in this invention; Figure 11 This is a schematic diagram of the circuit board of the power management energy storage module in the power management energy storage module of the present invention; In the diagram, 1 represents the heat sink fins; 2 is a multifunctional composite substrate, 24 is a double-sided copper-clad aluminum nitride ceramic substrate, 25 is a piezoelectric unit, 26 is a lower electrode layer, 27 is an upper electrode layer, 28 is a metal spring fixing seat, and 29 is an upper force-conducting ceramic substrate. 3 is the magnetic shielding protective shell, and 4 is the magnetic base; 5 is the electromagnetic vibration power generation module, 17 is the induction coil, 18 is the permanent magnet, and 19 is the elastic element; 6 is a thermoelectric power generation module, 20 is a P-type semiconductor, 21 is an N-type semiconductor, 22 is a copper current-carrying plate, and 23 is a ceramic substrate; 7 is the aviation connector, 8 is the housing of the multi-dimensional sensing device, 9 is the antenna, 10 is the 4G transmission module, 11 is the displacement sensor, 12 is the tilt sensor, and 13 is the high-performance, low-power microprocessor. 14 is the power management energy storage module circuit board, 15 is the lithium-ion battery, 16 is the power management energy storage module housing, and 30 is the supercapacitor. Detailed Implementation

[0018] The specific technical solution of the present invention will be further described in detail below with reference to specific examples.

[0019] As shown in the figure, the self-generating fault diagnosis and early warning system of the present invention includes a composite energy harvesting device, a power management energy storage module, a multi-dimensional sensing device, an edge computing diagnostic unit, and a remote monitoring platform. The system adopts a split electrical design. The composite energy harvesting device is used to convert the heat and mechanical vibration energy generated by the operation of the monitored equipment into electrical energy. It is essentially a type of thermo-engine dual-effect coupled generator, which is used to convert the waste heat and vibration energy generated by industrial equipment (such as motors, pumps, and pipelines) into electrical energy. After rectification, filtering, impedance matching, and steady-state processing by the power management energy storage module, it supplies power to the system load. The power management and energy storage module is used to regulate and store the electrical energy converted by the composite energy harvesting device, and to provide working power for the multi-dimensional sensing device and the edge computing diagnostic unit. The multidimensional sensing device is used to collect the displacement changes and triaxial acceleration changes of the monitored equipment in real time. It is integrated with the edge computing diagnostic unit and connected to the power management energy storage module through a flexible shielded line. The edge computing diagnostic unit has a built-in high-performance, low-power microprocessor 13, which is used to perform on-site analysis of data collected by multi-dimensional sensing devices, identify the fault characteristics of the monitored equipment, and predict its operating status in a future set time period. The edge computing diagnostic unit is also used to generate early warning information including fault probability, early warning level, expected fault type and fault level, and send the early warning information (including the fault probability, early warning level, expected fault type and corresponding fault level of the tested device in the future preset time period) to the remote monitoring terminal through the wireless communication module. The multi-dimensional sensing device and the edge computing diagnostic unit are integrated at a high density on the board level, and a low-loss electrical connection is established with the power management energy storage module through a flexible shielded circuit.

[0020] The composite energy harvesting device uses a longitudinally integrated stacked structure of thermoelectric-piezoelectric-electromagnetic to adapt to the complex and ever-changing energy environment in industrial sites; specifically, it includes heat dissipation plate fins 1, multi-functional composite substrate 2, magnetic shielding protective shell 3, magnetic base 4, electromagnetic vibration power generation module 5, and thermoelectric power generation module 6, etc. Among them, the electromagnetic vibration power generation module 5, the thermoelectric power generation module 6, and the multifunctional composite substrate 2 are responsible for energy supply; an electromagnetic shielding protective shell 3 is installed on the outer layer; the electromagnetic shielding protective shell 3 is made of high magnetic permeability alloy or die-cast aluminum alloy with conductive shielding layer, which constructs an electromagnetically compatible sealed cavity to shield the high-frequency electromagnetic radiation and electrostatic interference generated by external industrial equipment during operation. The entire device is attached to the surface of the industrial equipment being tested via a magnetic base 4 at the bottom, achieving non-invasive installation. It employs a dual-source power supply architecture of "thermoelectric-vibration" and a "thermo-mechanical-electric vertically integrated layered" manufacturing process, deeply integrating the three power generation mechanisms. The specific manufacturing process of this device is as follows: (1) The thermoelectric power generation module 6 serves as the basic energy source of the system. It converts the temperature difference between the waste heat and the cold end on the surface of the monitored equipment into continuous electrical energy and uses the waste heat generated during the stable operation of the equipment to continuously provide power. Specifically, it includes a P-type semiconductor 20, an N-type semiconductor 21, a copper current-conducting plate 22, and a ceramic substrate 23, etc.; it is set at the bottom as the substrate of the composite energy harvesting device. Its hot-end heat-conducting base is tightly attached to the high-temperature surface of the equipment through high thermal conductivity silicone grease, and a temperature difference field is established by utilizing the waste heat of the equipment operation. A continuous direct current is generated through the Seebeck effect of the PN junction array formed by the P-type semiconductor 20 and the N-type semiconductor 21; its core material is made of high-performance bismuth telluride (Bi2Te3) based thermoelectric material; its N-type semiconductor 21 is made of Bi2Te3 doped with 0.05wt% antimony iodide (SbI3). 2.7 Se 0.3 Solid solution, P-type semiconductor 20 uses Bi 0.5 Sb 1.5 Te3 alloys were used to optimize the thermoelectric figure of merit (ZT value) in the range of room temperature to 200°C. The preparation process is as follows: raw materials are mixed and melted according to stoichiometric ratio to synthesize polycrystalline ingots, which are then ball-milled and densified by spark plasma sintering (SPS). The ingots are then wire-cut into uniform grains, and a 1μm thick nickel barrier layer is electroplated on both ends of the grains to prevent solder diffusion. During packaging, a direct copper-coated alumina ceramic substrate 23 with high thermal conductivity is selected, and the grains are arranged in a π-type topology. The soldering is completed in a 280°C vacuum reflow oven using a high-temperature solder with a melting point of 240°C. (2) The multifunctional composite substrate 2 is both the cold end heat dissipation path of the thermoelectric power generation module 6 and the force-bearing base of the piezoelectric unit 15; it is the core component for realizing the decoupling and integration of thermoelectric power generation and vibration power generation. It has a double-sided functional structure. Its lower surface is a heat exchange surface, which is directly thermally coupled to the cold end surface of the thermoelectric power generation module through a thermally conductive interface material, serving as the heat dissipation path for thermoelectric power generation; the upper surface is an electromechanical conversion surface; and the piezoelectric unit 25 is attached thereon. A magnetic base 4 is also installed at the bottom of the multifunctional composite substrate 2. The multifunctional composite substrate 2, as the core coupling component connecting the temperature difference and vibration system, is installed above the cold end ceramic substrate 23 of the thermoelectric power generation module 6. It uses the periodic reaction force transmitted by the upper elastic element 19 to drive its own deformation, and then uses the piezoelectric effect of the attached material to convert mechanical strain energy into electrical energy. The main body of the substrate is a double-sided copper-clad aluminum nitride ceramic substrate 24 with a thickness of 1mm. The cold end of the thermoelectric power generation module 6 and the multifunctional composite substrate 2 share the same double-sided copper-clad aluminum nitride ceramic substrate 24 to take into account both thermal conductivity and mechanical elasticity. The lower surface of the substrate is bonded to the cold end of the temperature difference module through a high thermal conductivity interface material, which serves as a heat dissipation path. The copper layer on the upper surface of the double-sided copper-clad aluminum nitride ceramic substrate 24 serves as the lower electrode layer 26 of the multifunctional composite substrate 2. The piezoelectric unit 25 and the upper electrode layer 27 are coaxially attached above it. An upper conductive ceramic layer is insulated and attached above the upper electrode layer 27. The core material of the piezoelectric unit 25 is modified lead zirconate titanate (Pb(Zr, Ti)O3) piezoelectric ceramic (ring or disc structure lead zirconate titanate (PZT) piezoelectric ceramic sheet). For normal operating conditions, PZT-5H material is selected to maximize charge output; for high-temperature operating conditions (>120°C), PZT-8 material is selected to prevent thermal depolarization. At the geometric center of the upper surface of the substrate, a metal spring fixing seat 28 is fixed by laser welding process to rigidly connect the upper elastic element 19, thereby receiving the vibration reaction force and driving the substrate to generate piezoelectric deformation, and finally generating a voltage difference between the upper electrode layer 27 and the lower electrode layer 26. (3) The electromagnetic vibration power generation module 5 is located in the central cavity of the device, serving as an enhanced supplementary energy source for the system, providing high-power-density pulsed power during equipment startup, load fluctuations, or strong vibration conditions; it includes an induction coil 17, a permanent magnet 18, and an elastic element 19. A permanent magnet 18 is installed at the top of the elastic element 19. When the equipment vibrates, the permanent magnet 18 moves relative to the induction coil 17, cutting magnetic field lines to generate electromagnetic energy. At the same time, the periodic reaction force generated by the elastic element 19 acts directly on the center of the multifunctional composite substrate 2, forcing the substrate to drive the piezoelectric unit 25 to deform and output piezoelectric energy. Its bottom end is rigidly connected to the metal spring fixing seat 28. An induction coil 17 is installed around the permanent magnet 18 and the elastic element 19. It and the piezoelectric unit 25 on the multifunctional composite substrate 2 serve as supplementary energy to capture dynamic mechanical energy. The electromagnetic vibration power generation module 5 uses the mechanical vibration of the equipment during operation to drive the internal permanent magnet 18 and the induction coil 17 to perform relative cutting motion, thereby generating electricity.

[0021] The system utilizes the reaction force principle of a spring-mass system: when the permanent magnet 18 above vibrates violently under external excitation to generate an induced current (electromagnetic power generation), the elastic element 19 at the bottom supporting the magnet will generate a periodic pushing and pulling force of the same frequency on the substrate. This force directly drives the piezoelectric ceramic sheet integrated on the upper surface of the substrate to deform (piezoelectric power generation). This design converts the mechanical stress that was originally absorbed and dissipated by the base into electrical energy without increasing the size of the device, thus achieving full-frequency, dual-mechanism capture of vibration energy.

[0022] The three types of electrical energy from the electromagnetic vibration power generation module 5, the thermoelectric power generation module 6, and the multifunctional composite substrate 2 are combined and regulated by the power management energy storage module to form a power supply relationship that serves as a backup for each other.

[0023] The electromagnetic vibration power generation module 5 adopts an electromagnetic induction resonant structure; its core magnetic circuit uses a neodymium iron boron (NdFeB) permanent magnet 18 with a remanence Br≥1.45T, and the surface is treated with nickel-copper-nickel anti-corrosion plating. The induction coil 17 is made of polyimide enameled copper wire with a wire diameter of 0.05mm. It is wound on a nylon skeleton and then vacuum impregnated and dried to improve the seismic resistance. The elastic element 19 is made of beryllium bronze (QBe2) and undergoes precision stamping and vacuum aging hardening heat treatment to eliminate internal stress and improve fatigue life. During the manufacturing process, the weight of the mass block of the central oscillator is adjusted by laser to ensure that its natural frequency falls precisely within the main frequency range of the monitored equipment (such as 48Hz-52Hz), thereby maximizing power generation efficiency by utilizing the principle of resonance. (4) In terms of structural design, the magnetic shielding protective shell 3 encapsulated on the outside of the composite energy harvesting device (thermal difference power generation module 6, multifunctional composite substrate 2 and electromagnetic vibration power generation module 5) is used to isolate external interference. A quick-release slot is provided on the side wall of the magnetic shielding protective shell 3. The quick-release slot is used for modularly embedding heat sink fins 1. The heat sink fins 1 are made of aluminum alloy or copper graphite composite material. The fin spacing is optimized by simulation using Comsol software to enhance the natural convection heat transfer effect and ensure that sufficient thermoelectric temperature difference can be maintained even in windless environment. It is detachably installed on the magnetic shielding protective shell 3 through the slot to realize modular replacement and maintenance of heat dissipation components, so as to maintain sufficient temperature gradient between the hot and cold ends of the thermal difference power generation module even in windless environment in industrial site. (5) The magnetic base 4 installed at the bottom of the device adopts a detachable high-temperature resistant neodymium iron boron magnet. The contact surface can be designed as an arc structure that adapts to the curvature of the surface of the monitored equipment. It can adaptively adsorb onto the surface of pipes or housings with different curvatures. The magnetic force is sufficient to resist strong vibrations in the industrial field and prevent the sensor from falling off. This magnetic locking structure completely eliminates the need for drilling and hot welding processes, achieving true "on-demand" non-destructive deployment. It also provides great operational redundancy for subsequent non-stop inspection and relocation. The material of the lead zirconate titanate piezoelectric ceramic sheet is selected according to the surface temperature conditions of the monitored equipment. When the surface temperature of the equipment is below 120℃, PZT-5H material with a high voltage coefficient should be selected. When the surface temperature of the device is higher than 120°C, PZT-4 or PZT-8 modified materials with high Curie temperature are selected. The multifunctional composite substrate 2, which serves as the cold end of the thermoelectric power generation module, is made of a double-sided copper-clad ceramic plate with high thermal conductivity and high elasticity. Its operating temperature is lower than the depolarization temperature of the piezoelectric material.

[0024] The double-sided copper-clad aluminum nitride ceramic substrate 24 has thermal conductivity and insulation properties; The piezoelectric unit 25 is a lead zirconate titanate piezoelectric ceramic sheet with a ring or disc structure, which is coaxially attached to the upper surface of the lower electrode layer 26 by conductive silver paste or low-temperature sintering process.

[0025] The power management energy storage module adopts a hybrid energy storage architecture of "supercapacitor + lithium-ion battery" and is integrated inside the device. It includes a power management energy storage module shell 16, and a power management energy storage module circuit board 14, a lithium-ion battery 15 and a supercapacitor 30 encapsulated inside the power management energy storage module shell 16. The supercapacitor 30 is integrated on the power management energy storage module circuit board 14. This module is electrically connected to the composite energy harvesting device and the multi-dimensional sensing device through shielded cables. It integrates a three-channel impedance matching circuit and a hybrid energy storage architecture to rectify, boost and stabilize the collected unstable temperature difference energy and vibration energy, and prioritize the storage in the supercapacitor for use by the system load. At the same time, the lithium-ion battery 15 is used as a backup energy source under extreme conditions to ensure the uninterrupted operation of the system around the clock. The lithium-ion battery 15 is fixed to one side of the power management energy storage module circuit board 14 and electrically connected to the circuit board through wires. The power management energy storage module circuit board 14 leads out to the external interface through the aviation socket of the power management energy storage module housing 16 via shielded wires. The power management energy storage module adopts a multi-channel impedance matching input circuit and a hybrid energy storage architecture of supercapacitor 30 and lithium-ion battery 15. The power management energy storage module is designed with independent rectification and buck-boost circuits for thermoelectric power generation (low voltage, high current), electromagnetic power generation (medium voltage, medium impedance), and piezoelectric power generation (high voltage, high impedance). The supercapacitor 30 is used to buffer the pulse energy generated by vibration power generation and to provide instantaneous current for the wireless communication module to transmit data; The lithium-ion battery 15 serves as a backup power source to maintain the system's low-power sleep mode and timed heartbeat function when the monitored equipment is shut down for an extended period and there is no temperature difference or vibration (no heat or vibration).

[0026] The power management energy storage module is electrically connected to the composite energy harvesting device, and its circuit architecture adopts the mode of "three-channel impedance matching + hybrid energy storage". The power management energy storage module is designed with independent three-channel conditioning circuits for the output characteristics of the three energy sources (thermoelectric power generation - low voltage and high current, electromagnetic vibration power generation - medium voltage and medium impedance, and piezoelectric power generation - high voltage and high impedance). The two circuits are combined and stored in the supercapacitor to ensure the power supply continuity of the system under different operating conditions. Specifically: (1) Temperature difference path: Thermoelectric energy is collected by a boost converter chip with ultra-low start-up voltage; (2) Electromagnetic path: For the low voltage and low impedance AC current generated by the induction coil, a voltage doubler rectifier bridge is used for rectification; (3) Piezoelectric path: For the high voltage and high impedance AC power generated by the piezoelectric ceramic sheet, a full-bridge rectification and synchronous charge extraction circuit is used to maximize power transmission; After the three energy sources converge, they are stored in the supercapacitor 30 as the main buffer unit, which is responsible for absorbing the high-frequency pulse energy of vibration power generation and providing power support for the instantaneous high current when the wireless communication module transmits data. The lithium-ion battery 15 serves as a long-term backup energy source, which only intervenes in extreme cases where the equipment is shut down for a long time (without heat or vibration) and the supercapacitor is depleted, to maintain the system's low-power sleep timer and heartbeat packet transmission, thereby achieving lifelong maintenance-free operation.

[0027] The multi-dimensional sensing device includes an internally installed 4G transmission module 10, displacement sensor 11, tilt sensor 12, and high-performance low-power microprocessor 13. It continuously tracks the absolute displacement dimension and three-axis spatial acceleration of the target under test using multi-source information fusion technology. It cross-verifies the displacement collected by displacement sensor 11 and the three-axis acceleration collected by tilt sensor 12 to distinguish complex faults that are difficult to identify by a single sensor, thereby improving the accuracy of diagnosis and assisting the edge computing diagnostic unit in identifying complex faults. A multi-dimensional sensing device housing 8 is installed on the outside of the 4G transmission module 10, displacement sensor 11, tilt sensor 12 and high-performance low-power microprocessor 13. An aviation connector 7 and an antenna 9 are respectively installed on the outer wall of the multi-dimensional sensing device housing 8. The whole device is rigidly fixed to the surface of the industrial equipment under test by screws. A magnetic layer can be added to the bottom of the multi-dimensional sensing device housing 8 to assist in the reinforcement of the installation.

[0028] The multidimensional sensing device supports bidirectional wireless communication and is used to receive configuration commands issued by the remote monitoring terminal to adjust the sampling frequency, alarm threshold, or update the diagnostic model. The housing 8 of the multidimensional sensing device is made of die-cast aluminum alloy or high-strength engineering plastic, and is equipped with multiple vulcanized silicone sealing ring grooves. The joints are provided with double-layer silicone rubber sealing rings, which can resist oil stains, acid and alkali corrosion and dust erosion for a long time, ensuring the reliability of the internal precision circuits in harsh industrial environments.

[0029] The displacement sensor 11, tilt sensor 12, and 4G transmission module 10 in the multi-dimensional sensing device are arranged together with the high-performance, low-power microprocessor 13 in the edge computing diagnostic unit on the same PCB circuit board by surface mounting. They are directly fixed in the internal cavity of the composite energy harvesting device, reducing external wiring and connectors, significantly improving the system's shock resistance and electromagnetic compatibility, and reducing the overall size, making it easy to deploy in narrow spaces.

[0030] Specifically, the tilt sensor 12 is integrated on a PCB circuit board and uses an industrial-grade high-precision triaxial MEMS accelerometer chip, featuring low noise density and high temperature stability, supporting multiple range configurations of ±2g, ±4g, and ±8g. This sensor is used to collect the triaxial vibration acceleration and static tilt angle of the equipment in real time, with a resolution of up to 0.01°. In terms of sampling strategy, it supports dual-mode adaptive switching: during the stable operation period of the equipment, a low-power monitoring mode (sampling rate 10Hz-50Hz) is used to monitor trends; when the vibration amplitude is detected to exceed a set threshold, it automatically switches to a high-frequency diagnostic mode (sampling rate 1kHz-20kHz) to capture fault impact characteristics. The displacement sensor 11 adopts a non-contact detection principle and uses a miniature laser triangular reflective sensor. The displacement sensor 11 is integrated on a PCB circuit board, directly facing the surface of the equipment under test, with a repeatability accuracy better than 1mm, and is used to monitor the axial movement or thermal expansion displacement of the equipment rotor in real time.

[0031] Based on this, the multi-dimensional sensing device also has the ability to adaptively adjust the detection cycle by combining the power supply status and the equipment status. Specifically, the multi-dimensional sensing device obtains the remaining voltage value V and the voltage change rate dV / dt across the supercapacitor 30 in real time through the feedback path between the multi-dimensional sensing device and the power management energy storage module, so as to characterize the current energy reserve level and the net energy inflow / outflow rate of the system. At the same time, the vibration amplitude A is extracted after baseline removal of the displacement data collected by the displacement sensor 11, and the vibration frequency f is obtained by frequency domain analysis of the triaxial acceleration data collected by the tilt sensor 12, so as to characterize the current operating status of the monitored equipment.

[0032] The above four parameters (V, dV / dt, A, f) are combined for adaptive decision-making regarding the detection cycle: (1) When the energy is sufficient and the equipment is in abnormal condition: switch the detection cycle to the minimum value (e.g., 40μs, corresponding to a sampling rate of 25.6 kHz) and enter the high-frequency diagnostic mode to capture the transient characteristics of the fault to the greatest extent. (2) Sufficient energy and stable equipment status: The detection cycle is maintained at a normal value (e.g., 10ms-100ms), and it operates in a low-power standby mode; (3) Medium energy: Based on the deviation of vibration amplitude A and vibration frequency f from the normal reference, the detection cycle is dynamically adjusted by linear interpolation between the minimum value and the normal value to avoid blindly sampling at high frequency and depleting the stored energy; (4) Energy is limited or rapidly depleted: the detection cycle is extended to the maximum value (e.g., 1s-10min), and only the heartbeat packet transmission and over-limit interruption wake-up functions are retained to avoid monitoring blind spots; Through the adaptive adjustment mechanism of the above-mentioned "power status-equipment status" dual-factor coupling, this system can always maintain a dynamic balance between sampling density and energy availability in complex field conditions where micro-energy inputs such as temperature difference and vibration fluctuate with the working conditions. When abnormal signs appear in the equipment, it automatically increases the sampling density to capture fault characteristics, and automatically reduces the sampling density when energy reserves are tight to ensure that the system does not go offline.

[0033] The edge computing diagnostic unit incorporates a high-performance, low-power microprocessor 13, running a fault diagnosis and early warning algorithm based on "VMD + Deep Temporal Prediction Network + SVM". It can directly perform VMD variational mode decomposition on the original vibration signal locally to filter noise, extract multi-domain feature values, and input them into the SVM support vector machine model for comparison and matching. When abnormal monitoring data occurs, the system can automatically analyze and output the fault probability, early warning level, expected fault type, and fault level of the measured object within a future period, while simultaneously calculating the fault confidence. When the monitoring indicators exceed the safety threshold or match a severe fault mode, the system immediately triggers a high-priority alarm mechanism, thereby achieving millisecond-level fault response and qualitative and quantitative diagnosis, avoiding the early warning lag and cloud analysis pressure caused by network latency in the traditional "data pass-through" mode. This algorithm is used to perform time-frequency domain analysis on sensor data, identify fault characteristics such as imbalance, looseness, and wear of the equipment, and send early warning information to the remote monitoring terminal through a low-power wireless communication module (including but not limited to 4G network, LoRa, or NB-IoT). Its specific operation process is as follows: 1. Data preprocessing and slicing: The system first performs detrending processing on the acquired raw triaxial acceleration and displacement signals, and uses a sliding window mechanism to divide the continuous signal into several analysis samples, and uses overlapping sampling to increase the sample density; 2. Adaptive signal decomposition: Using the variational mode decomposition algorithm, the number of modes K is adaptively determined by the center frequency observation method, and the non-stationary signal is decomposed into K bandwidth-limited intrinsic mode components, and the energy entropy of each IMF component is calculated. 3. Dynamic feature sequence construction: Extract the time-domain statistical features and frequency-domain features of each IMF component, and fuse the displacement deviation values ​​collected by the displacement sensor. After Z-score standardization, construct a multi-dimensional feature vector, and concatenate them in chronological order to form a time series feature matrix. 4. Multi-step prediction of future trends: Input the time series feature matrix into the built-in deep time series prediction network, set the span of multi-step prediction by modifying the HORIZON parameters, and deduce and output the predicted feature vector of the corresponding future time node. 5. Fault Mode and Probability Assessment: The predicted feature vectors are input into a support vector machine classifier based on particle swarm optimization; The classifier uses a radial basis function kernel to output the posterior probability of each fault type occurring in the future, i.e., the fault probability, and determines the fault level based on the magnitude of the deviation of the feature vector from the normal baseline. 6. Intelligent Early Warning Fusion Decision: Introducing Dempster evidence theory, the output fault probability and the state of temperature and tilt sensors are mapped to a basic probability assignment function. The final fault confidence is calculated using Dempster synthesis rules. Combining the fault confidence and fault level, alarm messages including fault probability, early warning level, fault type and fault level are generated.

[0034] In addition, the wireless communication module in this invention adopts a 4G transmission module 10, which supports bidirectional communication. It can upload diagnostic results and early warning information, and can also receive configuration commands issued by the remote monitoring terminal for remotely adjusting the sampling frequency of the sensor, alarm threshold, or updating the diagnostic algorithm model. Under normal monitoring conditions, the system only sends heartbeat packets containing the device health index periodically, operating with extremely low power consumption. Once the edge computing diagnostic unit confirms that a fault has occurred, the communication module is immediately awakened and sends an emergency alarm frame with the highest priority. The frame data includes the fault code, the time of occurrence, and a short snapshot of the original waveform before and after the fault, for remote maintenance personnel to review.

[0035] The system core adopts a strict strong and weak current partitioning and shielding design principle. In the three-dimensional physical space, the high current transducer acquisition stage and the microvolt signal processing stage are forcibly isolated. In addition, a grounded metal shield tightly encloses the most sensitive sampling module and the high-performance low-power microprocessor 13, which prevents the severe electromagnetic surges in the industrial field from swallowing up the weak fault precursor characteristics, and prevents the strong electromagnetic interference in the industrial field from annihilating the weak fault signal characteristics.

[0036] This system can be installed on critical bridges, thermal pipelines, bearings, or elevator traction machines. When the equipment is running, the device utilizes waste heat and vibration for self-generation and self-use. When the equipment is running normally, the system monitors silently. Once an abnormal fault occurs inside the equipment, the edge computing unit will capture the abnormal change in the characteristic frequency within milliseconds and distinguish the fault type, realizing a leap from "passive inspection" to "proactive intelligent operation and maintenance".

[0037] Example 1: An application method of a self-generating fault diagnosis and early warning system in the health monitoring of bridge cables and main beam structures, comprising the following steps: First, the composite energy harvesting device is installed at the bottom of the bridge's steel box girder or in the cable anchorage area. Given the characteristics of the bridge environment, such as strong vibration, high wind speed, and high solar intensity, the system adopts a power supply strategy of "primarily using vibration energy and secondarily using temperature difference energy." The low-frequency vibration (0.5-20Hz) of the bridge caused by vehicle traffic and the wind-induced vibration directly drive the electromagnetic vibration power generation module 5 and the piezoelectric unit 25 to work, generating a large current as the main power source. Meanwhile, the temperature difference between the steel structure and the ambient air caused by sunlight drives the temperature difference power generation module 6 to generate trickle charging, which serves as an auxiliary power source to maintain the system in standby mode. The system is set to an event-triggered mode with a sampling frequency of 50Hz. Secondly, the multi-dimensional sensing device starts working; the laser probe of the displacement sensor 11 is aligned with the reference point of the main beam to monitor the vertical deflection change of the main beam of the bridge in real time; the tilt sensor 12 uses the internally integrated MEMS accelerometer to simultaneously monitor the triaxial vibration acceleration and static tilt angle of the tower or pier. The two are combined to capture the dynamic response and static deformation of the bridge structure in all directions. Subsequently, the data enters the edge computing diagnostic unit; the high-performance, low-power microprocessor 13 reads the current deflection, tilt angle, and vibration data, and performs multi-dimensional feature fusion analysis: Structural damage diagnosis: The VMD algorithm is used to decompose the triaxial acceleration signal and extract the natural frequency and damping ratio characteristics of the bridge structure. If the natural frequency drifts significantly (indicating a decrease in stiffness) and the deflection data shows irreversible residual deformation, it is determined that the bridge structure has fatigue damage or support detachment. Cable stress / attitude diagnosis: Establish a correlation model between deflection and tilt angle; if the deflection value displayed by displacement sensor 11 fails to return to zero after a vehicle passes, and tilt sensor 12 detects a slight change in the angle of the anchorage zone (indicating that the change in the cable stress state has led to local deformation), then the bridge is determined to be in a state of "cable slack" or "abnormal cable stress". Finally, the system outputs the diagnostic results; the edge computing diagnostic unit inputs the extracted feature vector into the fault diagnosis and early warning model, and outputs the probability of the bridge structure to experience the anomaly, the early warning level, the expected fault type and the fault level in the future; if "structural damage" or "cable force anomaly" is confirmed, the system immediately sends an alarm message containing the fault probability, early warning level, expected fault type, expected fault level and structural assessment suggestions to the remote monitoring terminal through the 4G transmission module 10, and triggers alarms on the local and remote monitoring terminals to notify the bridge maintenance engineer.

[0038] Example 2: An application method of a self-generating fault diagnosis and early warning system in monitoring the displacement and leakage of expansion joints in thermal pipelines, comprising the following steps: First, the composite energy harvesting device is attached to the fixed end flange of the thermal pipeline expansion joint; the fluid temperature inside the pipeline is 150-200℃, and the thermoelectric power generation module 6 uses the temperature difference between the waste heat on the pipeline surface and the environment to continuously generate a large current as the main energy source; the high-frequency micro-vibration generated by the fluid flow inside the pipeline drives the electromagnetic vibration power generation module 5 and the piezoelectric unit 25 to work as an auxiliary power source; the system is in an all-weather continuous monitoring mode, and the sampling interval is set to 10 minutes; Secondly, the multi-dimensional sensing device starts working; the laser probe of the displacement sensor 11 is aligned with the moving end of the pipe expansion joint to monitor the expansion and contraction displacement of the pipe expansion joint in real time; the tilt sensor 12 uses the internally integrated MEMS accelerometer to simultaneously monitor the triaxial vibration acceleration and static tilt angle of the pipe expansion joint. The two are combined to capture the running posture and motion characteristics of the pipe expansion joint in all directions. Subsequently, the data enters the edge computing diagnostic unit; the high-performance, low-power microprocessor 13 reads the current displacement, tilt angle, and vibration data, and performs multi-dimensional feature fusion analysis: Leak diagnosis: The VMD algorithm is used to decompose the high-frequency signal in the triaxial acceleration and extract the high-frequency energy entropy characteristics caused by fluid turbulence; if the high-frequency energy entropy increases suddenly and the displacement data does not change significantly, it is determined that the pipe expansion joint may have a rupture and leakage. Strain / Structural Diagnosis: Establish a correlation model between displacement and tilt angle; if the value of displacement sensor 11 remains static for a long time, but tilt sensor 12 detects a significant drift in the tilt angle of the flange plane (indicating that the pipeline has caused structural deflection or lateral deformation due to the inability to release stress caused by thermal expansion), then it is determined that the pipeline expansion joint is in a "stuck" or "skewed" state. Finally, the system outputs the diagnostic results; the edge computing diagnostic unit inputs the extracted feature vector into the fault diagnosis and early warning model, and outputs the probability of the bridge structure to experience the abnormality, the warning level, the expected fault type and the fault level in the future; if a "leak" or "expansion joint jamming" is confirmed, the system immediately sends an alarm message containing the fault probability, warning level, expected fault type, expected fault level and emergency disconnection suggestion to the remote monitoring terminal through the 4G transmission module 10, and triggers alarms on the local and remote monitoring terminals to remind on-site inspection personnel.

[0039] Example 3: An application method of a self-generating fault diagnosis and early warning system in the wear monitoring of rolling bearings in rotating machinery, comprising the following steps: First, the composite energy harvesting device is attached to the bearing housing surface of the non-drive end of a large rotating electric motor. Given the typical characteristics of the motor during operation—"continuous heating and high-frequency vibration"—the system adopts a dual-source energy supply strategy of "thermal energy difference and vibration energy working in tandem." The stable 60-80°C waste heat from the motor surface, combined with the heat dissipation fins, enables the thermoelectric power generation module 6 to continuously output a stable current as the basic energy supply. Simultaneously, the inherent mechanical vibration generated by the motor drives the electromagnetic vibration power generation module 5 and the piezoelectric unit 25 to operate, providing pulse power support for the system's high-frequency acquisition. The system is in high-frequency sampling and monitoring mode, with a sampling frequency of 25.6 kHz. Secondly, the multi-dimensional sensing device starts working; the laser probe of the displacement sensor 11 is aligned with the motor shaft extension end to monitor the radial runout of the shaft in real time; the tilt sensor 12 uses a MEMS accelerometer to simultaneously monitor the triaxial impact vibration of the bearing housing and the static levelness of the mounting base. The two are combined to capture the dynamic characteristics and assembly status of the rotating parts. Subsequently, the data enters the edge computing diagnostic unit, where a high-performance, low-power microprocessor 13 reads and analyzes the current jitter, tilt, and vibration data. Bearing damage diagnosis: The VMD algorithm is used to decompose the impact component in the acceleration signal and extract the fault characteristic frequency in the envelope spectrum. If obvious overtone components appear in the envelope spectrum and the displacement data shows that the radial runout of the shaft increases irregularly (indicating that the bearing clearance is increasing), it is determined that the rolling bearing has pitting, spalling or wear. Loosening / Misalignment Diagnosis: Establish a correlation model between radial runout and foundation tilt angle; if the radial runout value displayed by displacement sensor 11 fluctuates periodically, and the tilt sensor 12 detects a slight creep in the static tilt angle of the bearing housing over time (indicating that the foundation is unstable due to loose anchor bolts), then the equipment is determined to be in a state of "mechanical loosening" or "shaft misalignment". Finally, the system outputs the diagnostic results; the edge computing diagnostic unit inputs the extracted feature vector into the fault diagnosis and early warning model, and outputs the probability of the bridge structure to experience the abnormality, the warning level, the expected fault type and the fault level in the future; if "bearing wear" or "mechanical loosening" is confirmed, the system immediately sends an alarm message containing the fault probability, warning level, expected fault type, expected fault level and shutdown maintenance suggestions to the remote monitoring terminal through the 4G transmission module 10, and triggers alarms on the local and remote monitoring terminals to remind on-site inspection personnel.

[0040] Example 4: An application method of a self-generating fault diagnosis and early warning system in monitoring the operating status of elevator traction machines and cars, comprising the following steps: First, the composite energy harvesting device is installed on the elevator traction machine casing or car guide shoe bracket; the strong inertial force generated by the frequent start and stop of the elevator and the mechanical vibration of the running guide rail drive the electromagnetic vibration power generation module 5 and the piezoelectric unit 25 to work efficiently, serving as the main energy source of the system; at the same time, the waste heat emitted by the surface of the traction motor casing during long-term operation is utilized by the thermoelectric power generation module 6 as an auxiliary power source; the system is set to the "operation cycle" monitoring mode, that is, data analysis is performed once every time the elevator runs; Secondly, the multi-dimensional sensing device starts to work; the laser probe of the displacement sensor 11 is aligned with the shaft reference to monitor the elevator's leveling error and the relative extension and contraction of the wire rope in real time; the tilt sensor 12 uses the internally integrated MEMS accelerometer to simultaneously monitor the triaxial vibration acceleration (comfort index) and static levelness of the car or main unit. The two are combined to capture the elevator's running stability and mechanical structure status in all directions. Subsequently, the data enters the edge computing diagnostic unit; the high-performance, low-power microprocessor 13 reads the current displacement, tilt angle, and vibration data, and performs multi-dimensional feature fusion analysis: Wear / Comfort Diagnosis: The VMD algorithm is used to perform high-frequency decomposition on the Z-axis (vertical direction) acceleration signal to extract the high-frequency energy entropy characteristics generated by the friction between the guide shoe and the guide rail; if the high-frequency energy suddenly increases and the displacement data shows abnormal vibration during operation, it is determined that there is severe wear of the guide shoe or unevenness of the guide rail; Balance / load diagnosis: Establish a correlation model between leveling displacement and tilt angle; if the leveling error displayed by displacement sensor 11 continues to increase and tilt sensor 12 detects a fixed deflection angle of the base or car in a stationary state (indicating uneven load or abnormal counterweight), then the elevator is determined to be in a state of "uneven load" or "abnormal balance coefficient". Finally, the system outputs the diagnostic results; the edge computing diagnostic unit inputs the extracted feature vector into the fault diagnosis and early warning model, and outputs the probability of the bridge structure to experience the abnormality, the warning level, the expected fault type and the fault level in the future; if the "guide rail wear" or "abnormal off-center load" is confirmed, the system immediately sends an alarm message containing the fault probability, warning level, expected fault type, expected fault level and maintenance suggestions to the maintenance platform through the 4G transmission module 10, and triggers an audible and visual alarm locally to alert the maintenance personnel.

[0041] This invention utilizes a composite energy harvesting device attached to the wall of the target equipment to synchronously invert high-grade waste heat and broadband mechanical oscillations overflowing during the production cycle into clean electrical energy. After being regulated by a customized energy storage management circuit, this provides an inexhaustible power source for sensing and computing loads. The system employs a highly integrated electrical architecture, constructing an autonomous ecosystem of "energy flow capture—multi-dimensional sensing—edge inference—forward-looking alarm." Within this framework, the multi-dimensional sensing device integrates a displacement sensor 11 and a tilt sensor 12, which can accurately describe the structural deformation and dynamic anomaly characteristics. An edge computing diagnostic unit... Built-in intelligent algorithms perform on-site analysis and fault identification on the collected data, predict the probability of failure, warning level, fault type and fault level of the tested object in the future, and send the warning message containing the above information to the remote monitoring terminal through the wireless network. This invention constructs an intelligent early warning system with energy self-sufficiency capability, breaks through the power supply bottleneck and data transmission delay limitation of traditional monitoring methods, and has the characteristics of strong real-time performance, simple installation and maintenance, and all-weather operation. It shows irreplaceable effectiveness in the operation and maintenance of key equipment in petrochemical refining, power energy and metallurgical heavy industry.

Claims

1. A self-generating fault diagnosis and early warning system, characterized in that, include: The composite energy harvesting device is used to convert the thermal energy and mechanical vibration energy generated by the operation of the monitored equipment into electrical energy. The power management and energy storage module is used to regulate and store the electrical energy converted by the composite energy harvesting device, and to provide working power for the multi-dimensional sensing device and the edge computing diagnostic unit. The multi-dimensional sensing device is used to collect the displacement changes and triaxial acceleration changes of the monitored equipment in real time. It is integrated with the edge computing diagnostic unit and connected to the power management energy storage module through shielded wires. The edge computing diagnostic unit has a built-in high-performance, low-power microprocessor (13) for on-site analysis of data collected by the multi-dimensional sensing and detection device, identifying fault characteristics of the monitored equipment including component loosening, wear and structural cracks, pressure instability and liquid erosion, diagnosing fault probability, warning level, fault type and fault level, predicting the operating status of the monitored equipment in the future set time period to generate warning information, and sending it to the remote monitoring terminal through the wireless communication module.

2. The self-generating fault diagnosis and early warning system according to claim 1, characterized in that, The composite energy harvesting device includes, from bottom to top, a thermoelectric power generation module (6), a multifunctional composite substrate (2), and an electromagnetic vibration power generation module (5). The cold end of the thermoelectric power generation module (6) shares the same double-sided copper-clad aluminum nitride ceramic substrate (24) with the multifunctional composite substrate (2). Copper conductive plates (22) for soldering P-type semiconductors (20) and N-type semiconductors (21) are etched on the copper layer on the lower surface of the double-sided copper-clad aluminum nitride ceramic substrate (24). A ceramic substrate (23) is installed on the bottom side of the copper conductive plates (22) of the P-type semiconductors (20) and N-type semiconductors (21). The copper layer on the upper surface of the double-sided copper-clad aluminum nitride ceramic substrate (24) serves as the lower electrode layer (26) of the multifunctional composite substrate (2). A piezoelectric unit (25) and an upper electrode layer (27) are coaxially attached above it. An upper conductive ceramic substrate (29) is insulated and attached above the upper electrode layer (27). A metal spring fixing seat (28) is rigidly installed at the center position above the upper conductive ceramic substrate (29). The electromagnetic vibration power generation module (5) includes an elastic element (19) and a permanent magnet (18). The top of the elastic element (19) is equipped with a permanent magnet (18), and the bottom is rigidly connected to a metal spring fixing seat (28). An induction coil (17) is installed around the permanent magnet (18) and the elastic element (19), which together with the piezoelectric unit (25) on the multifunctional composite substrate (2) serve as a supplementary energy source. The multifunctional composite substrate (2) has a double-sided functional structure, with its lower surface being a heat exchange surface and its upper surface being an electromechanical conversion surface; a magnetic base (4) is also installed at the bottom of the multifunctional composite substrate (2). A magnetic shielding protective shell (3) is installed around the thermoelectric power generation module (6), the multifunctional composite substrate (2) and the electromagnetic vibration power generation module (5). A quick-release slot is provided on the side wall of the magnetic shielding protective shell (3), and a heat sink fin (1) is installed in the quick-release slot.

3. The self-generating fault diagnosis and early warning system according to claim 2, characterized in that, The magnetic base (4) is made of high temperature resistant rare earth permanent magnet material, and its contact surface is designed as an arc structure that adapts to the curvature of the surface of the monitored equipment. The double-sided copper-clad aluminum nitride ceramic substrate (24) has thermal conductivity and insulation properties; The piezoelectric unit (25) is a lead zirconate titanate piezoelectric ceramic sheet with a ring or disc structure, which is coaxially attached to the upper surface of the lower electrode layer (26) by conductive silver paste or low-temperature sintering process.

4. The self-generating fault diagnosis and early warning system according to claim 3, characterized in that, The material of the lead zirconate titanate piezoelectric ceramic sheet is selected according to the surface temperature conditions of the monitored equipment. When the surface temperature of the equipment is below 120℃, PZT-5H material with a high voltage coefficient should be selected. When the surface temperature of the equipment is higher than 120°C, PZT-4 or PZT-8 modified materials with high Curie temperature are selected, and the multifunctional composite substrate (2) serves as the cold end of the thermoelectric power generation module, with its operating temperature lower than the depolarization temperature of the piezoelectric material.

5. The self-generating fault diagnosis and early warning system according to claim 1, characterized in that, The power management energy storage module includes a power management energy storage module housing (16), a power management energy storage module circuit board (14) encapsulated inside the power management energy storage module housing (16), a lithium-ion battery (15), and a supercapacitor (30). The supercapacitor (30) is integrated on the power management energy storage module circuit board (14), the lithium-ion battery (15) is fixed on one side of the power management energy storage module circuit board (14) and electrically connected to the circuit board through a wire, and the power management energy storage module circuit board (14) leads out to the external interface through the aviation socket of the power management energy storage module shell (16) through a shielded wire.

6. The self-generating fault diagnosis and early warning system according to claim 1, characterized in that, The multidimensional sensing device supports bidirectional wireless communication and is used to receive configuration commands issued by the remote monitoring terminal to adjust the sampling frequency, alarm threshold, or update the diagnostic model. It includes a 4G transmission module (10), a displacement sensor (11), a tilt sensor (12), and a high-performance, low-power microprocessor (13) installed inside; a multi-dimensional sensing device housing (8) is installed outside the 4G transmission module (10), the displacement sensor (11), the tilt sensor (12), and the high-performance, low-power microprocessor (13). The housing (8) of the multidimensional sensing device is made of high-strength engineering plastic or die-cast aluminum alloy and is fitted with sealing rings at its joints. An aviation wiring plug (7) and an antenna (9) are respectively installed on its outer wall.

7. The self-generating fault diagnosis and early warning system according to claim 6, characterized in that, The multidimensional sensing device also has the function of adaptively adjusting the detection cycle, specifically including: (1) Read the remaining voltage value and voltage change rate in the power management energy storage module to reflect the energy conversion efficiency of the composite energy harvesting device; (2) Using the data collected by the displacement sensor (11) and the tilt sensor (12), calculate the current vibration amplitude and vibration frequency of the monitored object to reflect the operating status of the monitored object; Based on the above-mentioned energy status parameters and equipment status parameters, a joint judgment is made. When the energy is sufficient and the equipment shows abnormal signs, the detection cycle is shortened and high-frequency diagnosis is switched. When the energy is sufficient and the equipment status is stable, the regular detection cycle is maintained. When the energy reserves are tight, the detection cycle is extended to prioritize ensuring the system is online.

8. The self-generating fault diagnosis and early warning system according to claim 1, characterized in that, The displacement sensor (11), tilt sensor (12), and 4G transmission module (10) in the multidimensional sensing device, together with the high-performance low-power microprocessor (13) in the edge computing diagnostic unit, are all arranged on the same PCB circuit board by surface mount soldering.

9. The self-generating fault diagnosis and early warning system according to claim 1, characterized in that, The edge computing diagnostic unit operates based on a fault diagnosis and early warning algorithm, and includes the following steps: Step (1) Data preprocessing and slicing: The acquired raw triaxial acceleration and displacement signals are detrended, and the continuous signal is divided into several analysis samples using a sliding window mechanism. Overlap sampling is used to increase the sample density. Step (2) Adaptive signal decomposition: Using the variational mode decomposition algorithm, the number of modes K is adaptively determined by the center frequency observation method, and the non-stationary signal is decomposed into K bandwidth-limited intrinsic mode components, and the energy entropy of each IMF component is calculated; Step (3) Dynamic feature sequence construction: Extract the time domain statistical features and frequency domain features of each IMF component respectively, and fuse the displacement deviation values ​​collected by the displacement sensor. After Z-score standardization, construct a multi-dimensional feature vector and splice it in chronological order to form a time series feature matrix. Step (4) Multi-step prediction of future trends: Input the time series feature matrix into the built-in deep time series prediction network, set the span of multi-step prediction by modifying the HORIZON parameters, and deduce and output the prediction feature vector of the corresponding time node in the future. Step (5) Fault mode and probability assessment: Input the predicted feature vector into a support vector machine classifier based on particle swarm optimization; The classifier uses a radial basis function kernel to output the posterior probability of each fault type occurring in the future, i.e., the fault probability, and determines the fault level based on the magnitude of the deviation of the feature vector from the normal baseline. Step (6) Intelligent early warning fusion decision: Introduce DS evidence theory, map the output fault probability and the state of temperature and tilt sensors to the basic probability allocation function, and use Dempster synthesis rules to calculate the final fault confidence; combine fault confidence and fault level to generate alarm messages including fault probability, early warning level, fault type and fault level.

10. The working method of the self-generating fault diagnosis and early warning system as described in any one of claims 1-9, characterized in that, Includes the following steps: Step (1) Multidimensional data acquisition: The displacement change and triaxial acceleration change of the monitored equipment are acquired in real time through a multidimensional sensing device; Step (2) Edge-side local diagnosis: The edge computing diagnosis unit reads the collected data and runs the fault diagnosis and early warning algorithm locally; firstly, it uses VMD variational mode decomposition to denoise and extract features from the original signal to construct a multi-dimensional feature vector; then, it inputs the feature vector into the built-in deep temporal prediction network in time sequence to deduce the predicted feature vector of the monitored device in the future set time period; then, it uses the built-in SVM support vector classifier to identify fault modes, and finally outputs the posterior probability of each fault type occurring in the future period. Step (3) Intelligent early warning decision: The edge computing diagnostic unit introduces the DS evidence theory, integrates the posterior probability output by SVM with the auxiliary sensing status, calculates the final fault confidence, and makes local decisions accordingly: if the device is determined to be operating normally, the wireless communication module is controlled to remain in sleep mode or only send heartbeat packets. If the equipment is determined to be faulty and the confidence level exceeds the set threshold, an early warning message is generated, including the fault probability, warning level, fault type, and fault level. Step (4) Remote wireless transmission: The edge computing diagnostic unit wakes up the wireless communication module and sends the warning information to the remote monitoring terminal; the remote monitoring terminal receives and displays the warning information for maintenance personnel to view.

Images