Cold storage dynamic defrosting control and fault early warning system based on AI algorithm

Abstract

The invention relates to the technical field of refrigeration equipment control, and discloses a cold storage dynamic defrosting control and fault early warning system based on an AI algorithm, and the system comprises a sensing detection assembly and a central processing unit. The central processing unit superposes the micro-disturbance signal into an electronic expansion valve control instruction to drive the valve to generate pressure pulsation excitation; and synchronously acquiring stator current, air return pressure, air return temperature and valve opening signals of the air cooler. The processing unit calculates a thermodynamic residual characteristic value representing the heat exchange performance, an aerodynamic impedance characteristic value representing the aerodynamic load, and a cross-domain coherence characteristic value of the return air pressure and the stator current at the excitation frequency. Based on the multi-dimensional characteristics, the system recognizes that the evaporator is in a frostless, loose-frost, compact-ice or non-frosting type air duct blocking state at present, and performs defrosting, operation maintaining or alarming operation according to the current state of frostless, loose-frost, compact-ice or non-frosting type air duct blocking state. According to the invention, through active excitation and cross-domain feature fusion, quantitative distinguishing of frost layer texture and accurate early warning of non-frosting faults are realized.

Description

Technical Field

[0001] This invention relates to the field of refrigeration equipment control technology, specifically to a dynamic defrosting control and fault early warning system for cold storage based on AI algorithms. Background Technology

[0002] In cold storage refrigeration systems, forced convection heat transfer using evaporators (coolers) is a common method for maintaining a low-temperature environment inside the storage room. Due to the low-temperature operating conditions, frost formation on the evaporator fins is an unavoidable physical phenomenon. As the refrigeration system operates, frost gradually accumulates on the evaporator surface, increasing thermal resistance, reducing the heat exchange efficiency between the refrigerant and air, and also occupying fin gaps, increasing airflow resistance and leading to a decrease in airflow. Therefore, timely and accurate defrosting is crucial for ensuring refrigeration performance and reducing system energy consumption.

[0003] Existing defrosting control strategies mainly include timed defrosting and on-demand defrosting based on a single physical parameter. Timed defrosting strategies do not consider changes in actual ambient humidity and cargo throughput, and are prone to performing defrosting in frost-free or slightly frosty conditions, resulting in energy waste and unnecessary fluctuations in storage temperature; or failing to defrost in time when frost is severe, leading to compressor operational risks. Control methods based on a single thermodynamic parameter (such as pipe temperature, temperature difference, superheat) mainly focus on the attenuation of heat exchange capacity. However, the physical properties of the frost layer change dynamically during its formation, gradually evolving from an initial loose and porous frost layer to a later dense ice layer. Loose frost layers and dense ice layers differ significantly in thermal resistance and mechanical properties, and it is difficult to effectively distinguish between the two based solely on thermodynamic parameters. This often leads to the system initiating defrosting prematurely when the frost layer is still in a loose and permeable stage, or responding only after the ice layer has become excessively dense, causing "ice blockage."

[0004] On the other hand, to monitor airflow blockage, some existing technologies have introduced fan current or pressure difference detection methods. While these methods can reflect changes in duct resistance, their signals lack specificity. When the air inlet of the evaporative cooler is blocked by packaging materials, or when fan blades are damaged, causing load changes, the signal characteristics fed back by the sensors are easily confused with severe frosting conditions. Existing control systems lack comprehensive identification capabilities across physical fields, making it difficult to distinguish between gradual flow channel blockage caused by frosting and non-frosting blockage caused by foreign object obstruction or equipment failure. In the event of non-frosting blockage, the system often incorrectly identifies it as severe frosting and forcibly executes a defrosting procedure. This not only fails to solve the physical blockage problem but also raises the storage temperature due to heating operations, posing a potential threat to the safety of cold chain goods. Summary of the Invention

[0005] This invention aims to solve the technical problems in existing cold storage defrosting control technologies, such as the difficulty in distinguishing between loose frost layers and dense ice layers, and the difficulty in identifying non-frost-forming air duct blockages.

[0006] This invention provides a dynamic defrosting control and fault early warning system for cold storage based on AI algorithms by introducing active micro-perturbation excitation and cross-physical field signal analysis technology.

[0007] The first aspect of the present invention provides a dynamic defrosting control and fault early warning system for cold storage based on AI algorithms. The system is deployed in a refrigeration cycle device including a compressor, condenser, electronic expansion valve, evaporator and air cooler.

[0008] The system includes a sensing and detection component and a central processing unit.

[0009] The sensing and detection components are configured to collect the stator current signal of the air cooler, the return gas pressure signal of the refrigerant, the return gas temperature signal, and the valve opening signal of the electronic expansion valve.

[0010] The central processing unit is communicatively connected to the sensing and detection components and the electronic expansion valve, and is configured to perform the following operations: superimpose micro-perturbation signals onto the control commands of the electronic expansion valve to drive the electronic expansion valve to generate pressure pulsation excitation in the refrigerant pipeline; calculate thermodynamic residual characteristic values ​​characterizing heat transfer performance based on return gas pressure signals, return gas temperature signals, and valve opening signals; calculate aerodynamic impedance characteristic values ​​characterizing aerodynamic load based on stator current signals; calculate the frequency domain correlation between return gas pressure signals and stator current signals at the pressure pulsation excitation frequency, and extract cross-domain coherence characteristic values; identify the physical state of the evaporator based on thermodynamic residual characteristic values, aerodynamic impedance characteristic values, and cross-domain coherence characteristic values, and output control commands accordingly.

[0011] In one alternative implementation, the central processing unit generates a sinusoidal waveform with a preset frequency and amplitude as a micro-perturbation signal, and drives the electronic expansion valve to perform periodic reciprocating adjustment movements near the current base opening position. This micro-perturbation signal serves as an active excitation source for system identification, used to establish traceable dynamic characteristic components in the refrigerant fluid.

[0012] In one alternative implementation, in order to improve the signal-to-noise ratio of signal analysis, the central processing unit performs envelope extraction processing on the stator current signal to obtain the current envelope signal before calculating the aerodynamic impedance characteristic value and the cross-domain coherence characteristic value, and performs bandpass filtering processing on the return gas pressure signal and the return gas temperature signal to retain the dynamic components near the pressure pulsation excitation frequency.

[0013] In one optional implementation, the calculation process of the cross-domain coherence eigenvalue includes: calculating the cross-power spectral density of the filtered return gas pressure signal and the current envelope signal; calculating the amplitude squared coherence function based on the cross-power spectral density; extracting the amplitude of the amplitude squared coherence function at the frequency of the micro-disturbance signal as the coherence coupling degree eigenvalue, and extracting the phase difference between the return gas pressure signal and the current envelope signal as the phase lag angle eigenvalue. These coherence coupling degree eigenvalues ​​and phase lag angle eigenvalues ​​can quantify the mechanical coupling efficiency and transmission delay of pressure fluctuations transmitted to the fan load end through the pipeline structure and frost layer medium.

[0014] In one optional implementation, the calculation process of the thermodynamic residual characteristic value includes: calculating the actual observed total heat transfer coefficient based on the lumped-parameter thermodynamic model using the return gas pressure signal, return gas temperature signal, and valve opening signal; and calculating the normalized difference between the actual observed total heat transfer coefficient and the preset theoretical frost-free heat transfer coefficient. This characteristic value reflects the attenuation of heat transfer performance caused by frost as a thermal resistance.

[0015] In one optional implementation, the calculation of the aerodynamic impedance characteristic value includes: calculating the power spectral density of the stator current signal envelope; and integrating the power spectral density over the rotor slip frequency sideband. This characteristic value reflects the change in fan load torque caused by frost clogging the air duct.

[0016] In one optional implementation, the central processing unit (CPU) identifies the physical state of the evaporator by constructing a multi-dimensional feature vector, which includes thermodynamic residual eigenvalues, aerodynamic impedance eigenvalues, coherent coupling eigenvalues, and phase lag angle eigenvalues. The CPU calculates the weighted Euclidean distance between the multi-dimensional feature vector and a preset standard state centroid vector, and selects the state corresponding to the minimum distance as the determination result. The standard state centroid vector corresponds at least to a frost-free state, a loose frost accumulation state, and a dense ice-locked state. When calculating the weighted Euclidean distance, the weighting coefficients of the coherent coupling eigenvalues ​​and the phase lag angle eigenvalues ​​are set to be greater than the weighting coefficients of the thermodynamic residual eigenvalues ​​and the aerodynamic impedance eigenvalues.

[0017] In one optional implementation, the central processing unit performs hierarchical control based on the identification results: when the identification result is a dense ice lock-in state, it outputs a defrosting start command; when the identification result is a loose frost accumulation state, it outputs a maintain operation command and shortens the injection cycle of the micro-disturbance signal to enter a high-frequency monitoring mode; when the identification result is a non-frost-type air duct blockage state, it outputs a fault alarm command.

[0018] A second aspect of this invention provides a method for dynamic defrosting control and fault early warning in cold storage based on AI algorithms, comprising the following steps: The micro-disturbance signal is superimposed on the control command of the electronic expansion valve to drive the electronic expansion valve to generate pressure pulsation excitation in the refrigerant pipeline; Collect the stator current signal, refrigerant return gas pressure signal, refrigerant return gas temperature signal, and electronic expansion valve valve opening signal of the air cooler; Based on the return gas pressure signal, return gas temperature signal and valve opening signal, the thermodynamic residual characteristic value characterizing the heat transfer performance is calculated. Based on the stator current signal, the aerodynamic impedance characteristic value characterizing the aerodynamic load is calculated; Calculate the frequency domain correlation between the return gas pressure signal and the stator current signal at the pressure pulsation excitation frequency, and extract cross-domain coherence feature values; The physical state of the evaporator is identified based on thermodynamic residual characteristic values, aerodynamic impedance characteristic values, and transdomain coherence characteristic values, and control commands are output accordingly.

[0019] This invention provides a dynamic defrosting control and fault early warning system for cold storage based on AI algorithms. It has the following beneficial effects: 1. This invention employs active micro-perturbation excitation technology. By injecting pressure pulsations of a specific frequency at the electronic expansion valve and detecting the response of these pulsations at the fan current terminal, a system identification channel spanning the fluid and electrical domains is constructed. This cross-domain coherence analysis can directly detect the dynamic coupling characteristics of the mechanical link between the evaporator piping, frost layer, air, and fan blades. When the frost layer evolves from a loose state to a dense ice layer, its structural stiffness and medium damping change significantly, leading to changes in the transmission efficiency and phase lag of pressure fluctuations. This method overcomes the limitations of traditional single thermodynamic or single electrical parameter detection, achieving quantitative differentiation of the physical texture of the frost layer (loose frost or dense ice).

[0020] 2. This invention integrates three dimensions of physical characteristics: thermodynamic residual, aerodynamic impedance, and cross-domain coherence. Thermodynamic residual characterizes heat transfer efficiency, aerodynamic impedance characterizes duct resistance, and cross-domain coherence characterizes structural and media properties. The fusion of these multi-dimensional characteristics significantly improves the robustness of state recognition, particularly in effectively distinguishing between conventional frosting and non-frosting duct blockages (such as foreign object obstruction or fan malfunction), thus avoiding false defrosting or missed alarms in the system.

[0021] 3. The present invention adopts a hierarchical control strategy. When loose frost accumulation is detected, the micro-disturbance injection cycle is shortened to enter the high-frequency monitoring mode, which can accurately capture the critical point of frost phase change or flow channel blockage, thereby realizing on-demand defrosting and reducing unnecessary defrosting energy consumption and storage temperature fluctuations. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the structure of a cold storage dynamic defrosting control and fault early warning system based on AI algorithm according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating an AI-based dynamic defrosting control and fault early warning method for cold storage, according to an embodiment of the present invention. Detailed Implementation

[0023] To more clearly illustrate the technical solutions of the embodiments in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some examples or embodiments of this specification. For those skilled in the art, these drawings can be applied to other similar scenarios without creative effort. Unless obvious from the context or otherwise specified, the same reference numerals in the drawings represent the same structures or operations.

[0024] See attached document Figure 1 , Figure 1 This is a schematic diagram of a dynamic defrosting control and fault early warning system for cold storage based on an AI algorithm, according to an embodiment of the present invention. The present invention provides a dynamic defrosting control and fault early warning system for cold storage based on an AI algorithm. This system is deployed in a refrigeration cycle unit including a compressor, condenser, electronic expansion valve, evaporator, and air cooler. The system mainly includes a sensing and detection component, a central processing unit, and an execution control component.

[0025] The sensing and detection components are configured to acquire operating status data of the refrigeration system, and include a current acquisition unit, a pressure sensor, and a temperature sensor. The current acquisition unit is connected in series in the power supply circuit of the evaporator and is configured to acquire the high-frequency stator current signal of the evaporator. The pressure sensor is installed in the return gas line of the evaporator and is configured to acquire the refrigerant return gas pressure signal. The temperature sensor is installed in the return gas line of the evaporator and is configured to acquire the refrigerant return gas temperature signal. The electronic expansion valve has a communication interface and is configured to provide feedback on the current valve opening signal.

[0026] The central processing unit (CPU) is connected to the sensing and detection components and the execution and control components, and internally runs AI state recognition algorithms and signal processing logic. The CPU includes a micro-perturbation excitation injection module, a signal synchronous acquisition module, a signal preprocessing module, a thermodynamic residual calculation module, an aerodynamic impedance analysis module, a cross-domain coherence analysis module, a state decision module, and a control execution module.

[0027] The micro-perturbation excitation injection module is connected to the drive port of the electronic expansion valve. The micro-perturbation excitation injection module is configured to generate a micro-perturbation signal containing a sinusoidal waveform of a specific frequency, and superimpose the micro-perturbation signal into the basic control command of the electronic expansion valve to drive the electronic expansion valve to generate periodic pressure pulsation excitation in the refrigerant pipeline.

[0028] The signal synchronization acquisition module is connected to the current acquisition unit, pressure sensor, temperature sensor and electronic expansion valve respectively; the signal synchronization acquisition module is configured to synchronously receive and store stator current signal, return gas pressure signal, return gas temperature signal and valve opening signal, and perform timestamp alignment on each signal.

[0029] The signal preprocessing module is connected to the signal synchronous acquisition module. The signal preprocessing module is configured to perform Hilbert transform on the stator current signal to extract the current envelope signal, and to perform bandpass filtering on the return gas pressure signal and return gas temperature signal to retain the dynamic components near the excitation frequency.

[0030] The thermodynamic residual calculation module is connected to the signal preprocessing module. The thermodynamic residual calculation module is configured to calculate the actual heat transfer coefficient of the evaporator based on the lumped parameter thermodynamic model, and calculate the theoretical frost-free heat transfer coefficient in combination with the benchmark model. The thermodynamic residual characteristic value is output by comparing the difference between the two.

[0031] The aerodynamic impedance analysis module is connected to the signal preprocessing module. The aerodynamic impedance analysis module is configured to calculate the power spectral density of the stator current envelope signal, integrate to calculate the energy value in the rotor slip frequency sideband, and output the aerodynamic impedance characteristic value.

[0032] The cross-domain coherence analysis module is connected to the signal preprocessing module. The cross-domain coherence analysis module is configured to calculate the cross-power spectral density and coherence function of the return gas pressure signal and the stator current envelope signal at the excitation frequency, and extract the coherence coupling characteristic value and the phase lag angle characteristic value.

[0033] The state decision module is connected to the thermodynamic residual calculation module, the aerodynamic impedance analysis module, and the cross-domain coherence analysis module. The state decision module is configured to receive the above feature values ​​and construct a multi-dimensional feature vector, and input the vector into a preset state classifier for logical operation to identify the current physical state of the evaporator.

[0034] The control execution module is connected to the status decision module. The control execution module is configured to output corresponding control commands based on the identified physical status, including defrost start commands, fault alarm commands, or maintenance operation commands.

[0035] See attached document Figure 2 , Figure 2 This is a flowchart illustrating a method for dynamic defrosting control and fault early warning of cold storage based on an AI algorithm, according to an embodiment of the present invention. The present invention provides a method for dynamic defrosting control and fault early warning of cold storage based on an AI algorithm, comprising the following steps: S100, the micro-perturbation excitation injection module generates a sinusoidal micro-perturbation signal with a preset frequency and amplitude, and drives the electronic expansion valve to generate pressure pulsation excitation in the refrigerant pipeline; S200, the signal synchronization acquisition module synchronously acquires the stator current signal of the air cooler, the return gas pressure signal of the evaporator, the return gas temperature signal, and the valve opening signal of the electronic expansion valve. S300, the signal preprocessing module performs envelope extraction processing on the stator current signal and filtering processing on the return gas pressure signal and return gas temperature signal; S400, the central processing unit performs parallel computation of multidimensional physical characteristics: The thermodynamic residual calculation module calculates the thermodynamic residual characteristic values ​​that characterize the decay of heat transfer performance. The aerodynamic impedance analysis module calculates the aerodynamic impedance characteristic values ​​that characterize the aerodynamic load; The cross-domain coherence analysis module calculates the characteristic values ​​of coherent coupling degree and phase hysteresis angle, which characterize the signal transmission properties. S500's state decision module constructs a feature vector based on thermodynamic residual feature values, aerodynamic impedance feature values, coherent coupling feature values, and phase lag angle feature values. It then uses a state classifier to determine whether the current operating condition is frost-free, loose frost accumulation, dense ice blockage, or non-frost-forming duct blockage. S600, the control execution module executes the control strategy according to the judgment result: if it is determined to be a dense ice lock-in state, it performs defrosting operation; if it is determined to be a loose frost accumulation state, it maintains cooling operation; if it is determined to be a non-frost type air duct blockage state, it outputs an alarm signal.

[0036] See attached document Figure 1 The physical architecture of the AI-based cold storage dynamic defrosting control and fault early warning system is built upon a typical vapor compression refrigeration cycle. This refrigeration cycle mainly consists of a compressor, condenser, electronic expansion valve, and evaporator, all connected in sequence. A cooling fan is positioned adjacent to the evaporator or integrated with it as a single unit, used to force airflow through the evaporator's heat exchange fins.

[0037] In terms of hardware connectivity, the current acquisition unit is connected in series to the stator power supply circuit of the air cooler motor, or it is mounted on the stator power supply cable as a non-intrusive current transformer. To meet the requirements of the subsequent aerodynamic impedance analysis module for extracting the rotor slip frequency sideband component and fluid turbulence noise, the current acquisition unit is configured as a high-frequency acquisition device. Its sampling frequency must meet the requirements of the Nyquist sampling theorem for the frequency passing through the motor rotor bars and higher harmonics. In specific implementations, the current acquisition unit can use a high-precision Hall current sensor or a shunt, converting the analog current signal into a digital time-series signal. The signal is transmitted to the central processing unit. This stator current signal contains characteristic components modulated onto the fundamental current by the aerodynamic load torque fluctuations.

[0038] The pressure and temperature sensors are physically installed at the return gas port of the evaporator, specifically on the pipe section between the evaporator outlet and the compressor suction port. The pressure sensor is a piezoresistive or capacitive pressure transmitter configured to monitor the refrigerant return gas pressure in real time. This pressure signal is responsive to the pipeline pressure wave caused by the action of the electronic expansion valve. The temperature sensor is a resistance temperature detector (RTD) (such as PT100) or a thermocouple; the probe is inserted into the pipeline or installed close to the pipe wall, configured to detect the refrigerant return temperature. The installation, selection, and signal transmission circuit design of pressure and temperature sensors are well-known technologies to those skilled in the art and will not be elaborated upon here.

[0039] The electronic expansion valve, serving as both a throttling element and an actuator for micro-disturbance excitation in the system, is driven by a stepper motor. In addition to receiving drive pulses from the central processing unit, the electronic expansion valve also features position feedback or a communication interface (such as RS485 Modbus), configured to provide real-time feedback of the current valve opening signal to the central processing unit. This opening signal serves as reference data for the control domain, used for synchronous analysis with signals from the thermal and electrical domains.

[0040] The central processing unit (CPU) establishes data connections with the current acquisition unit, pressure sensor, temperature sensor, and electronic expansion valve via analog input interfaces (A / D converters) or fieldbus interfaces. The CPU is the core of the system's computation and can be implemented as a digital signal processor (DSP), field-programmable gate array (FPGA), or embedded microcontroller (MCU). The CPU integrates a micro-perturbation excitation injection module, a signal synchronization acquisition module, a signal preprocessing module, a thermodynamic residual calculation module, aerodynamic impedance analysis module, cross-domain coherence analysis module, state decision module, and control execution module. These modules are functional units stored in memory as computer program instructions and executed by the processor, or hardware logic units implemented as application-specific integrated circuits (ASICs). The CPU also connects to the refrigeration system's start-stop control loop to output defrosting commands, alarm signals, and equipment start-stop signals.

[0041] The micro-perturbation excitation injection module, located within the central processing unit, generates excitation signals with specific frequency characteristics. This module introduces active micro-perturbations into the control loop of the electronic expansion valve, producing pressure pulsation signals in the refrigerant fluid that can be recognized by subsequent algorithms. These signals serve as a phase reference for subsequent cross-domain signal analysis. The specific operational logic of the micro-perturbation excitation injection module includes the following steps: S201, Generate Basic Superheat Control Command. The central processing unit reads the return gas pressure signal and return gas temperature signal, calculates the current superheat value, and calculates the basic opening command of the electronic expansion valve required to maintain the target superheat based on a preset closed-loop control algorithm. This basic opening command is denoted as... Its main function is to maintain the thermodynamic balance of the refrigeration system and the stability of the storage temperature. The above closed-loop control algorithm can adopt a proportional-integral-derivative (PID) control algorithm or a fuzzy logic control algorithm. The specific parameter tuning and implementation logic of such algorithms are well known in this field and will not be elaborated here.

[0042] S202, Construct a sinusoidal micro-perturbation excitation signal and a synchronization reference. The micro-perturbation excitation injection module generates a continuous time-series sinusoidal signal, denoted as... The mathematical model of this signal is defined as follows: ; In the formula, This represents the value of the micro-perturbation excitation signal at time t; It represents the amplitude of the micro-perturbation excitation, and its unit is consistent with the opening unit (steps or percentage) of the electronic expansion valve; The frequency of the micro-perturbation excitation is expressed in Hertz (Hz); t represents the time variable. Indicates the initial phase.

[0043] Regarding the selection of the above parameters, amplitude The value range is limited to 1% to 3% of the full stroke of the electronic expansion valve. The lower limit of this amplitude setting must be greater than the mechanical dead zone or gear hysteresis of the electronic expansion valve's stepper motor to ensure substantial displacement of the valve needle; the upper limit must be less than the thermal disturbance threshold that causes significant fluctuations in the storage temperature. Frequency The set value avoids the operating frequency of the refrigeration compressor, the pass-through frequency (BPF) of the air cooler blades and its harmonics, and the mechanical resonance frequency range of the piping system. In this embodiment, the frequency... The frequency range is set to a low frequency range of 0.05Hz to 0.1Hz. This frequency band is lower than the cutoff frequency of the time constant corresponding to the thermal inertia of the evaporator, but higher than the cutoff frequency of fluid pipeline transmission. This ensures that enough complete periodic waveforms can be acquired within a short time window.

[0044] Meanwhile, the micro-perturbation excitation injection module will generate signal or its frequency With phase The parameters serve as a synchronization reference signal and are transmitted in real time to the cross-domain coherence analysis module for subsequent related demodulation operations.

[0045] S203, Synthesized Composite Drive Command. The micro-perturbation excitation injection module injects the basic opening command. With sinusoidal micro-perturbation excitation signal Perform linear superposition operations to generate the final composite drive command sent to the electronic expansion valve. The composition operation is expressed as follows: ; This composite drive instruction It includes a DC component for maintaining operating conditions and an AC component for detecting frost characteristics.

[0046] S204 executes the physical field excitation generation. The electronic expansion valve responds to the composite drive command. The valve needle is driven to perform a slight reciprocating adjustment movement near its current basic opening position. This adjustment movement periodically changes the flow cross-sectional area of ​​the refrigerant through the valve port, causing the refrigerant mass flow rate and velocity entering the evaporator to fluctuate at the same frequency. Under the action of pressure difference, this flow fluctuation is transformed into a flow with a specific frequency within the closed pipeline. The pressure pulsation waves and fluid turbulence noise. These pressure pulsations, which are actively introduced by the control end and whose frequency and phase are precisely known, constitute the source excitation signal required for subsequent cross-domain coherence analysis.

[0047] The signal synchronization acquisition module and signal preprocessing module are configured to perform time alignment and feature enhancement processing on multi-source heterogeneous data to address the issues of inconsistent sampling frequencies from different physical field sensors and low signal-to-noise ratio of the original signal. The processing flow of this part specifically includes the following steps: S301, Synchronization and Downsampling Adaptation for Multi-rate Signals. The signal synchronization acquisition module establishes a global time alignment mechanism based on the system's unified master clock reference. To accommodate both low-frequency micro-perturbation analysis (0.01Hz) and mid-frequency aerodynamic impedance analysis (slip sideband approximately 1-10Hz), the module sets a uniform analysis sampling rate. The frequency is 50Hz. This takes into account the original sampling frequency of the current acquisition unit (denoted as ). For frequencies significantly higher than the analysis sampling frequency (e.g., 10kHz), this module employs an anti-aliasing downsampling strategy to adapt to subsequent cross-domain coherence analysis. Specifically, for high-frequency stator current signals... After completing the envelope extraction in step S302, the system proportionally extracts the high-frequency envelope signal to a value similar to... Consistent low-frequency sampling rate; for return gas pressure signals Return gas temperature signal and electronic expansion valve opening feedback signal The module maintains its original sampling rate and uses timestamp indexes to perform point-to-point alignment with the downsampled current envelope signal to construct a state data matrix of a unified dimension.

[0048] S302, Hilbert envelope extraction of stator current. The signal preprocessing module processes the raw stator current signal. Demodulation is performed to extract the baseband signal reflecting the fluctuations in the wind turbine load torque. This step first inputs the raw signal to a power frequency bandpass filter. The center frequency of this filter is set to the fundamental frequency of the power grid (50Hz or 60Hz), and the passband width is set to cover the range of the maximum slip frequency of the motor rotor and the main sideband components (e.g., ...). ±5Hz) to filter out high-frequency harmonic interference and retain sideband information containing load characteristics.

[0049] Subsequently, an analytic signal is constructed using the Hilbert transform. The filtered current signal is defined as a real function. Its Hilbert transform Defined as and Convolution: ; Based on this, an analytic signal is constructed. Current envelope signal Obtained by calculating the magnitude of the analytic signal: ; In the formula, This represents the extracted instantaneous amplitude envelope of the current; Let be the measured current value at time t; The value of the Hilbert transform at time t; It is the integral variable.

[0050] S303, frequency domain filtering and feature preservation of thermal and control signals. Specifically for return gas pressure signals. Return gas temperature signal and valve opening signal The signal preprocessing module performs targeted filtering. This takes into account the micro-perturbation excitation frequency injected in step S200. Given the parameters, this module is configured with a digital bandpass filter (such as a second-order Butterworth bandpass filter), whose passband center frequency is locked at [value missing]. The bandwidth is set to ±0.2. This bandwidth setting is designed to filter out background thermal noise from the refrigeration system and fluid pulsations in unrelated frequency bands.

[0051] Filtered dynamic pressure components Represented as: ; In the formula, Indicates the bandpass filter operator; This is the original pressure signal; BW is the center value of the excitation frequency; BW is the passband width.

[0052] This step removes broadband high-frequency noise and extremely low-frequency temperature drift, retaining only the physical field response components related to the active excitation action of the electronic expansion valve. Furthermore, the thermodynamic residual calculation module obtains the DC steady-state component of the aforementioned signal through a parallel low-pass filter branch, which is used to calculate the static thermal resistance characteristics.

[0053] The central processing unit quantifies the current physical state of the evaporator from both thermodynamic and aerodynamic dimensions by running the thermodynamic residual calculation module and the aerodynamic impedance analysis module in parallel. This process aims to extract independent eigenvalues ​​reflecting the thermal resistance and flow resistance effects of frost layers. The specific calculation process includes the following steps: S401, constructs a real-time heat transfer coefficient observation model for the evaporator. The thermodynamic residual calculation module estimates the current cooling capacity and overall heat transfer coefficient using the lumped parameter method. This module first reads the opening feedback signal of the electronic expansion valve. , return air pressure signal and return gas temperature signal The module obtains the DC steady-state component. Simultaneously, it acquires current condensing pressure or exhaust pressure data through the communication interface between the central processing unit and the refrigeration unit. If measured values ​​are unavailable, a preset saturated pressure estimation model is used to obtain the value based on the current ambient temperature. The refrigerant mass flow rate is estimated based on the valve's flow characteristic equation. : ; In the formula, The valve flow coefficient; To match the valve opening The corresponding flow cross-sectional area function; The density of the refrigerant before the valve is determined by the discharge pressure. And the subcooling is calculated; This is the return air pressure.

[0054] Subsequently, the product of the observed total heat transfer coefficient and the area was calculated. : ; In the formula, This indicates the current observed heat transfer capacity; The refrigerant outlet specific enthalpy is obtained by looking up a table based on the return gas pressure and temperature; The specific enthalpy at the evaporator inlet is approximately taken as the enthalpy of the liquid before the valve. The temperature difference is the logarithmic mean.

[0055] S402 calculates the normalized thermodynamic residual characteristics. The thermodynamic residual calculation module internally stores a pre-calibrated benchmark heat transfer model. This benchmark model is constructed through experimental calibration using a multidimensional lookup table or polynomial fitting function, describing the theoretical overall heat transfer coefficient under clean (frost-free) evaporator surface conditions. This is a functional mapping relationship between the current fan speed and heat exchange temperature difference. The module inputs the current operating conditions into this function to calculate the baseline value. And calculate the thermodynamic residual eigenvalues. : ; In the formula, This is a dimensionless characteristic value. When frost forms on the evaporator surface, the frost layer acts as an additional thermal resistance, affecting the actual observed value. The decline caused It tends from 0 to 1.

[0056] S403 performs power spectral density analysis of the current envelope. The aerodynamic impedance analysis module receives the stator current envelope signal from the signal preprocessing module. Since changes in airflow resistance alter the load torque of the air cooler, thereby adjusting the amplitude of the stator current, this module uses the Welch method to calculate the power spectral density (PSD) of the current envelope signal, denoted as... .

[0057] S404, extracting aerodynamic impedance characteristic values. The aerodynamic impedance analysis module uses power spectral density... It locks onto a specific characteristic frequency band. This characteristic frequency band The range is set to cover the characteristic fault frequencies in the stator current envelope spectrum of the motor. and its sidebands, among which s is the rotor slip of the motor. This is the fundamental frequency of the power supply. When frost blocks the fin gaps, causing the back pressure in the airflow path to increase, the fan load torque increases, leading to an increase in slip s, which in turn significantly increases the spectral energy density within this frequency band.

[0058] aerodynamic impedance characteristic value The following results were obtained by integrating the power spectral density within the characteristic frequency band: ; In the formula, It represents the total energy value within a specific frequency band. This value is specifically used to characterize the aerodynamic impedance on the evaporator side and directly reflects the degree of blockage in the air-side flow channel.

[0059] The cross-domain coherence analysis module, as the core algorithm unit of this invention, is configured to perform system identification tasks between physical fields. This module quantifies the energy transfer characteristics between different physical fields by calculating the correlation between signals from the thermo-fluid domain and the electrical domain in the frequency domain. The specific analysis process includes the following steps: S501, constructing the input and output variables of the cross-physics system identification model and determining the analysis time window. The cross-domain coherence analysis module first defines the input signals of the system identification model. With output signal In this embodiment, the dynamic component of the return gas pressure output in step S303 is... Defined as system input signal This signal characterizes the fluid excitation source acting on the evaporator piping system; the stator current signal output in step S301... Defined as system output signal .

[0060] To meet the spectral resolution requirements of low-frequency excitation signals (e.g., 0.01Hz), the module sets the time window length for data acquisition and analysis. This length Configured to contain at least K complete micro-perturbation excitation cycles (where This ensures that subsequent Welch analyses have sufficient averaging times. The module... and Perform synchronous truncation to obtain a discrete time series of length N.

[0061] S502 performs piecewise windowing and discrete Fourier transform. To suppress spectral leakage and reduce the variance of random noise, the cross-domain coherence analysis module uses the Welch average periodogram method to process the data. First, the input sequence of length N is... and output sequence The data is divided into L segments of length M, with a 50% overlap between adjacent segments. Then, a Hanning or Hamming window is applied to each segment, and the Fast Fourier Transform algorithm is used to convert it into a frequency domain complex sequence.

[0062] No. The transformation formula for each data segment is as follows: ; ; In the formula, and The first The Fourier transform results of each data segment at the discrete frequency k; represents the window function coefficients; M represents the segment length.

[0063] S503 calculates the cross-power spectral density (CPSD) and the self-power spectral density (PSD). The cross-domain coherence analysis module approximates the power spectral density of the signal by averaging the periodograms of L data segments, thereby eliminating incoherent noise interference.

[0064] The self-power spectral density of the input signal and the auto-power spectral density of the output signal The calculation is as follows: ; ; Cross-power spectral density between input and output signals The calculation is as follows: ; In the formula, L is the average number of segments; The sampling frequency; The normalized power coefficient of the window function; This represents complex conjugation. The cross-power spectral density... The calculation results retain the phase difference information between the input and output signals at each frequency point.

[0065] S504, Calculate the amplitude-squared coherence function. Based on the averaged power spectral density data above, the cross-domain coherence analysis module calculates the amplitude-squared coherence function. And specifically extract the excitation frequency of micro-perturbation. The function value at that point. The calculation formula is as follows: ; In the formula, The coherence coefficient is dimensionless, and its value range is... When at the excitation frequency point When the coherence coefficient approaches 1, it indicates that the energy of the output signal (current fluctuation) is mainly caused by the linear relationship of the input signal (pressure fluctuation), that is, the refrigerant pressure pulsation is effectively transmitted to the fan load end; when it approaches 0, it indicates that the transmission path is blocked or is subject to strong noise interference.

[0066] Based on the frequency domain analysis results obtained from the preceding steps, the cross-domain coherence analysis module further executes feature extraction logic, aiming to extract quantitative indicators characterizing the mechanical properties of frost from the complex spectral data. The specific implementation steps for this part are as follows: S505, extracting coherent coupling characteristic values. The cross-domain coherence analysis module calculates the amplitude-squared coherence function spectrum. In the middle, the micro-perturbation excitation frequency is locked. The frequency point at which it is located. To eliminate the effects of spectral leakage and frequency discretization errors, the module does not simply take the value of the center frequency, but instead calculates the value of the frequency point at which it is located. The average energy within a narrow frequency window centered on the coherent coupling degree is used as a characteristic value. .

[0067] The calculation formula is as follows: ; In the formula, To correspond to the excitation frequency Discrete frequency index value; It is half the bandwidth (e.g., taking 1 to 3 frequency resolution units); This is the coherence function value at frequency index m.

[0068] This eigenvalue Physically, this characterizes the dynamic coupling efficiency of the mechanical transmission link between the evaporator piping, fins, air, and fan blades. Specifically, when the evaporator is in a frost-free or initial frosting state, the piping structure possesses a specific elastic modulus, and internal pressure pulsations can effectively induce microscopic radial deformation of the pipe wall, which is then transmitted to the fan load through the air medium, exhibiting high efficiency. Value. When the evaporator surface is covered with a dense ice layer, the additional stiffness of the ice layer restricts the radial elastic deformation of the tube wall, and the high mechanical impedance of the ice layer attenuates the transmission efficiency of vibration waves to the air side, resulting in a lower coherent coupling degree. Significant decrease.

[0069] S506, extract the relative phase lag eigenvalues. The cross-domain coherence analysis module utilizes the complex cross-power spectral density calculated in step S503. Extract the input signal (pressure excitation) and output signal (current response) at the excitation frequency. The phase difference at that point. To accurately obtain phase information for the entire cycle, the module uses a four-quadrant arctangent function to calculate the absolute phase angle and subtracts a preset system-native reference phase. The relative phase hysteresis caused solely by frost was obtained. : ; ; In the formula, The range of function return values ​​is It is the inherent transmission delay phase that is pre-calibrated and stored in the system under frost-free drying conditions.

[0070] This eigenvalue The group delay characteristics of mechanical waves propagating in a frosty medium were characterized. As a porous medium, variations in the thickness and density of the frosty layer alter the propagation velocity and damping ratio of the vibration waves, resulting in a phase lag angle through the medium. It exhibits a monotonic variation trend related to the frost layer structure.

[0071] S507, construct a normalized multidimensional physical state feature vector. The cross-domain coherence analysis module uses the coherent coupling degree feature values ​​calculated above. Relative phase lag eigenvalues The thermodynamic residual characteristic value obtained from the aforementioned calculation and aerodynamic impedance characteristic value Summarize the results.

[0072] To eliminate the influence of different physical dimensions on subsequent decision weights, the module employs a linear maximum-minimum normalization algorithm, mapping each eigenvalue to the [0,1] interval based on its physical limit boundary, thus constructing an eigenvector. : ; In the formula, , , , These are all system-preset physical boundary constants, corresponding to the system's clean state limit value and complete failure state limit value, respectively. For Although its theoretical range is [0,1], in practical engineering, due to signal-to-noise ratio limitations, its measured values ​​in clean and fully locked states are usually within specific intervals (e.g., [0.8,0.95] and [0.1,0.3]), thus requiring normalization to enhance feature discriminative power. This feature vector... Information from four independent physical dimensions—thermodynamic heat transfer efficiency, aerodynamic flow resistance, structural stiffness coupling characteristics, and medium transmission damping characteristics—is then transmitted to the state decision module.

[0073] The state decision module and control execution module are configured based on the normalized feature vectors generated in the aforementioned steps. The current evaporator operating condition is determined using a weighted minimum distance classifier, and corresponding control commands are output accordingly. This process is based on pattern recognition in a multi-dimensional feature space, and the specific implementation steps are as follows: S601, constructs the baseline centroid of the multi-dimensional state space. The non-volatile memory of the state decision module contains a pre-built state feature library. This feature library defines the cluster centers (centroids) of three standard operating conditions in the four-dimensional feature space, as follows: Clean state center of mass This corresponds to the operating conditions where the evaporator is frost-free, the airflow is unobstructed, and the tube wall and fins are well elastically coupled. Loose frosted core This corresponds to the evaporator surface being covered with a porous, loose frost layer, and its thermal resistance characteristics. Increased but mechanically coupled characteristics Operating conditions with minimal changes; Dense icy core This corresponds to the condition where the frost layer ages and densifies, forming ice bridges, which leads to flow channel blockage and the tube fin structure exhibits high-rigidity locking.

[0074] The above centroid vector All of them have the same The same dimensional structure. These centroid data were obtained by performing K-Means clustering analysis on the collected historical feature vector sequences through a full-cycle frosting experiment on a specific model of air cooler in a standard enthalpy difference laboratory, using fixed parameters.

[0075] S602, calculate the weighted Euclidean distance. This is to quantize the current state vector. The state decision module calculates the similarity between the state and the centroids of each standard state. With each center of mass Weighted Euclidean distance between The calculation formula is as follows: ; In the formula, n is the index of the characteristic dimension, which corresponds to thermodynamic residual, aerodynamic impedance, coherent coupling degree and relative phase lag, respectively; Let be the weight coefficient of the nth feature dimension, satisfying .

[0076] The system's preset weight allocation strategy sets the weight values ​​for coherent coupling and relative phase lag. , ) is greater than the weighting value of thermodynamic residual and aerodynamic impedance ( , This configuration allows the system to prioritize cross-domain coherence characteristics that reflect the mechanical properties of the medium in its decision-making logic, thereby distinguishing between loose frost and dense ice based on differences in mechanical stiffness under conditions where thermodynamic characteristics are not significant (such as low humidity environments).

[0077] S603, performs state arbitration and anomaly detection. The state decision module compares the three calculated distance values. , , The state corresponding to the minimum value is selected as the candidate judgment result. Meanwhile, to identify non-frost-related faults (such as fan blade breakage or sensor drift), the module introduces a confidence threshold. .

[0078] If minimum distance If this occurs, it indicates that the current state has deviated from all known normal evolution paths, and the system determines that it is currently in an "abnormal fault state".

[0079] like The system then locks the current state based on the minimum value index: When judged as When this happens, output status code S0; When judged as When this happens, output status code S1; When judged as When this happens, output status code S2.

[0080] S604 generates and executes hierarchical control strategies. The control execution module receives status codes output by the status decision module and drives peripheral devices to perform corresponding operations via I / O interfaces or bus communication. In response to status code S0 (clean), the control execution module maintains the current cooling control logic unchanged and instructs the micro-perturbation excitation injection module to continue sending probe signals at a standard preset cycle (e.g., once every 10 minutes).

[0081] In response to status code S1 (loose frost), the control execution module maintains refrigeration operation to ensure stable storage temperature, but immediately switches the micro-perturbation excitation injection module to "high-frequency monitoring mode" and shortens the injection cycle of the detection signal (e.g., adjusts it to once every 2 minutes) to track the critical inflection point of the phase change of the frost layer density in real time with high resolution.

[0082] In response to status code S2 (dense icing), the control execution module immediately triggers the defrosting procedure. Specific operations include: stopping the micro-disturbance excitation injection; closing the liquid supply solenoid valve and executing the pump set operation; closing the defrosting contactor or switching the four-way valve. Simultaneously, the system records the current defrosting trigger time and the corresponding feature vector. Upload the data to the historical database, which will be used by subsequent systems to analyze the centroid vector. Learning samples that undergo adaptive correction.

[0083] In response to an "abnormal fault state", the control execution module executes the safety shutdown logic and issues a fault alarm code through the human-machine interface to prompt maintenance personnel to check the equipment.

[0084] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A cold storage dynamic defrosting control and fault early warning system based on AI algorithms, characterized in that, The system includes: The sensing and detection component is configured to collect the stator current signal of the air cooler, the return gas pressure signal of the refrigerant, the return gas temperature signal, and the valve opening signal of the electronic expansion valve. The central processing unit is communicatively connected to the sensing and detection components and the electronic expansion valve, and is configured to perform the following operations: The micro-perturbation signal is superimposed on the control command of the electronic expansion valve to drive the electronic expansion valve to generate pressure pulsation excitation in the refrigerant pipeline; Based on the return gas pressure signal, return gas temperature signal and valve opening signal, calculate the thermodynamic residual characteristic value that characterizes the heat transfer performance; Based on the stator current signal, the aerodynamic impedance characteristic value characterizing the aerodynamic load is calculated; Calculate the frequency domain correlation between the return gas pressure signal and the stator current signal at the pressure pulsation excitation frequency, and extract cross-domain coherence feature values; The physical state of the evaporator is identified based on the thermodynamic residual characteristic value, aerodynamic impedance characteristic value, and transdomain coherence characteristic value, and control commands are output accordingly.

2. The AI-based dynamic defrosting control and fault early warning system for cold storage as described in claim 1, characterized in that, The central processing unit is configured to generate a sinusoidal waveform containing a preset frequency and amplitude as the micro-perturbation signal, and drive the electronic expansion valve to perform periodic reciprocating adjustment motion near the current basic opening position.

3. The cold storage dynamic defrosting control and fault early warning system based on AI algorithm according to claim 1, characterized in that, Before calculating the aerodynamic impedance characteristic value and the transdomain coherence characteristic value, the central processing unit is configured as follows: The stator current signal is subjected to envelope extraction processing to obtain the current envelope signal; The return gas pressure signal and return gas temperature signal are subjected to bandpass filtering to retain the dynamic components near the pressure pulsation excitation frequency.

4. The AI-based dynamic defrosting control and fault early warning system for cold storage as described in claim 3, characterized in that, The central processing unit is specifically configured as follows when calculating cross-domain coherence eigenvalues: Calculate the cross-power spectral density between the filtered return gas pressure signal and the current envelope signal; Calculate the amplitude squared coherence function based on the cross-power spectral density; The amplitude of the squared coherence function at the frequency of the micro-disturbance signal is extracted as the coherence coupling feature value, and the phase difference between the return gas pressure signal and the current envelope signal is extracted as the phase lag angle feature value.

5. The AI-based dynamic defrosting control and fault early warning system for cold storage as described in claim 1, characterized in that, The central processing unit is specifically configured as follows when calculating the thermodynamic residual characteristic values: Based on the lumped parameter thermodynamic model, the actual observed total heat transfer coefficient is calculated using the return gas pressure signal, return gas temperature signal and valve opening signal. Calculate the normalized difference between the actual observed total heat transfer coefficient and the preset theoretical frost-free heat transfer coefficient.

6. The cold storage dynamic defrosting control and fault early warning system based on AI algorithm according to claim 1, characterized in that, The central processing unit is specifically configured as follows when calculating the aerodynamic impedance characteristic value: Calculate the power spectral density of the stator current signal envelope; The aerodynamic impedance characteristic value is obtained by integrating the power spectral density within the rotor slip frequency sideband range.

7. The AI-based dynamic defrosting control and fault early warning system for cold storage as described in claim 4, characterized in that, When identifying the physical state of the evaporator, the central processing unit is configured as follows: Construct a multidimensional feature vector containing the aforementioned thermodynamic residual feature values, aerodynamic impedance feature values, coherent coupling feature values, and phase hysteresis angle feature values; Calculate the weighted Euclidean distance between the multidimensional feature vector and the preset standard state centroid vector; The state corresponding to the minimum distance is selected as the judgment result; the standard state centroid vector corresponds at least to the frost-free state, the loose frost accumulation state, and the dense ice closure state.

8. The cold storage dynamic defrosting control and fault early warning system based on AI algorithm according to claim 7, characterized in that, When calculating the weighted Euclidean distance, the weighting coefficients of the coherent coupling characteristic value and the phase lag angle characteristic value are set to be greater than the weighting coefficients of the thermodynamic residual characteristic value and the aerodynamic impedance characteristic value.

9. The AI-based dynamic defrosting control and fault early warning system for cold storage as described in claim 7, characterized in that, The central processing unit is configured as follows: When the identification result is a dense ice-locked state, a defrosting start command is output; When the identification result is a loose frost accumulation state, a maintenance operation command is output, and the injection cycle of the micro-perturbation signal is shortened to enter the high-frequency monitoring mode; When the identification result is a non-frost-forming air duct blockage, a fault alarm command is output.

10. The cold storage dynamic defrosting control and fault early warning system based on AI algorithm according to claim 1, characterized in that, The sensing and detection component includes: A current acquisition unit is connected in series in the power supply circuit of the air cooler and is configured as a high-frequency acquisition device; The pressure sensor and temperature sensor are physically installed on the pipe section between the evaporator outlet and the compressor suction port.

Images