Refrigerating unit fault analysis cloud service system and control method

Abstract

The application relates to the technical field of refrigeration equipment monitoring, and discloses a refrigerating unit fault analysis cloud service system and a control method. The method comprises a data acquisition module, a characteristic extraction module, a fault evaluation module, an analysis optimization module and a cloud service module. The data acquisition module acquires real-time refrigerating unit operation state data through a multi-source sensor; the characteristic extraction module receives the operation state data and extracts operation characteristic data; the fault evaluation module generates a characteristic baseline value according to the operation characteristic data, judges the operation state difference degree based on the baseline value and generates initial fault parameters; the analysis optimization module determines a parameter optimization direction according to the initial fault parameters, acquires a historical fault data set, matches the characteristic baseline value with the historical fault data set and generates a fault optimization signal; and the cloud service module receives the fault optimization signal and generates a cloud platform control instruction. The system realizes intelligent analysis and processing of refrigerating unit faults, and improves the accuracy and efficiency of fault management.

Description

Technical Field

[0001] This invention relates to the field of refrigeration equipment monitoring technology, specifically to a cloud service system and control method for refrigeration unit fault analysis. Background Technology

[0002] In modern industrial production and commercial operations, refrigeration units, as key equipment for maintaining specific temperature environments, are widely used in many fields such as food cold chain, pharmaceutical storage, and data center cooling. Their stable operation is directly related to the continuity of production processes, the reliability of product quality, and the controllability of operating costs. However, the working environment of refrigeration units is often quite complex, involving the coordinated operation of multiple core components such as compressors, condensers, and evaporators. During long-term operation, these components are susceptible to factors such as temperature fluctuations, pressure changes, and refrigerant leaks, leading to frequent failures.

[0003] Traditional methods of troubleshooting refrigeration units largely rely on manual inspections and experience-based judgment. Maintenance personnel must periodically visit the site to check the equipment's operating status, inferring potential faults by observing instrument data and listening to mechanical sounds. This method is not only labor-intensive and time-consuming, but also limited by the professional skills and experience of the personnel, making it difficult to accurately predict and promptly address potential faults. Often, faults are only discovered after they have already occurred and caused downtime, resulting in incalculable losses such as production interruptions and product damage.

[0004] With the development of sensor and IoT technologies, some refrigeration units have begun to be equipped with data acquisition devices, enabling them to acquire real-time parameters such as temperature, pressure, and current. However, most existing systems can only perform simple data storage and display, lacking the ability to deeply analyze and mine the data. They cannot extract characteristic information reflecting the health status of the equipment from massive amounts of operational data, let alone assess and predict faults based on this information. Furthermore, due to differences in the models and operating environments of different refrigeration units, the lack of unified fault judgment standards and baseline values ​​significantly reduces the accuracy of fault identification.

[0005] In troubleshooting, traditional systems struggle to effectively reuse historical fault data. When a new fault occurs, maintenance personnel often have to start from scratch, unable to quickly draw on past experience with similar faults, leading to low troubleshooting efficiency. Furthermore, because data is scattered across different devices, cross-regional and cross-device fault data sharing and collaborative analysis are impossible, limiting the overall improvement of fault handling capabilities. These problems mean that refrigeration unit fault management remains in a reactive state, failing to meet the high reliability and stability requirements of modern industrial production. Summary of the Invention

[0006] The purpose of this invention is to provide a cloud service system and control method for refrigeration unit fault analysis, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a cloud service system for refrigeration unit fault analysis, the system comprising:

[0008] Data acquisition module: Acquires real-time operating status data of the refrigeration unit through multi-source sensors;

[0009] The feature extraction module receives the operating status data and extracts the operating characteristic data of the refrigeration unit.

[0010] Fault assessment module: Generates characteristic baseline values ​​based on the operating characteristic data of the refrigeration unit; determines the degree of difference in operating status based on the characteristic baseline values; and generates initial fault parameters based on the degree of difference in operating status.

[0011] Analysis and optimization module: Determines the parameter optimization direction based on the initial fault parameters; acquires a historical fault data set based on the parameter optimization direction; matches the feature baseline value with the historical fault data set, and generates a fault optimization signal based on the matching result;

[0012] The cloud service module receives the fault optimization signal and generates cloud platform control commands.

[0013] Preferably, receiving the operating status data and extracting the chiller unit operating characteristic data includes:

[0014] Obtain the compressor vibration spectrum, condensing temperature time-series curve, and evaporating pressure change rate from the operating status data;

[0015] The vibration spectrum of the compressor is decomposed in the frequency domain to obtain the set of main frequency amplitudes;

[0016] The condensation temperature time series curve is smoothed to obtain a stable temperature sequence.

[0017] Calculate the unit-time fluctuation of the evaporation pressure change rate;

[0018] The main frequency amplitude set, stable temperature sequence, and unit time fluctuation are integrated into the operating characteristic data of the refrigeration unit.

[0019] Preferably, generating the characteristic baseline value based on the refrigeration unit operating characteristic data includes:

[0020] Extract the historical average amplitude of the main frequency during the same period from the operating characteristic data of the refrigeration unit;

[0021] Obtain the fluctuation threshold of the stable temperature sequence;

[0022] Calculate the historical normal range of the fluctuation per unit time;

[0023] The historical average amplitude of the dominant frequency, the fluctuation threshold, and the historical normal range are weighted and fused to generate a characteristic baseline value.

[0024] Preferably, determining the degree of difference in operating status based on the feature baseline value includes:

[0025] The amplitude offset is obtained by calculating the difference between the current dominant frequency amplitude and the average dominant frequency amplitude of the same period in history;

[0026] The duration for which the stable temperature sequence exceeds the fluctuation threshold is detected;

[0027] Determine whether the fluctuation per unit time deviates from the historical normal range;

[0028] The operational state difference is generated based on the amplitude offset, duration, and departure state.

[0029] Preferably, generating initial fault parameters based on the difference in operating states includes:

[0030] The difference in operating status is compared with a preset difference level threshold;

[0031] When the difference in operating status reaches the first difference level threshold, compressor frequency adjustment parameters are generated as initial fault parameters.

[0032] When the difference in operating status reaches the second difference level threshold, refrigerant flow correction parameters are generated as initial fault parameters.

[0033] When the difference in operating status reaches the third difference level threshold, system shutdown detection parameters are generated as initial fault parameters.

[0034] Preferably, matching the feature baseline value with the historical fault data set includes:

[0035] Search the historical fault data set for a target historical feature that is the same as the feature baseline value;

[0036] When a single target historical feature exists, retrieve the historical optimization record corresponding to that target historical feature;

[0037] When multiple target historical features exist, calculate the average optimization coefficient of the multiple target historical features;

[0038] When no target historical features exist, the feature similarity calculation process is initiated.

[0039] Preferably, the initiation feature similarity calculation process includes:

[0040] Calculate the characteristic distance between the characteristic baseline value and each historical fault data;

[0041] Filter valid historical data where the feature distance is less than the distance threshold;

[0042] The percentage of valid historical data is calculated.

[0043] The feature matching weight is determined based on the stated proportion of quantities.

[0044] Preferably, generating the fault optimization signal based on the matching result includes:

[0045] When historical optimization records are obtained, the historical optimization records are used to generate the first optimization instruction signal;

[0046] When the average optimization coefficient is obtained, the average optimization coefficient is fused with the current initial fault parameters to generate a second optimization command signal;

[0047] When the feature matching weight is obtained, the initial fault parameters are adjusted according to the feature matching weight to generate a third optimization instruction signal;

[0048] The fault optimization signal includes a first optimization command signal, a second optimization command signal, or a third optimization command signal.

[0049] Preferably, receiving the fault optimization signal and generating cloud platform control commands includes:

[0050] Upon receiving the first optimization command signal, a compressor frequency recalibration command is generated;

[0051] When the second optimization command signal is received, a refrigerant flow stage adjustment command is generated.

[0052] When the third optimization instruction signal is received, a system retest instruction is generated;

[0053] The cloud platform control commands drive the refrigeration unit to perform corresponding operations.

[0054] Preferably, the present invention also includes a cloud service control method for refrigeration unit fault analysis, the method comprising all modules and method flow of the refrigeration unit fault analysis cloud service system described above.

[0055] Compared with the prior art, the beneficial effects of the present invention are:

[0056] Through the collaborative operation of multiple modules, intelligent management of the entire process of refrigeration unit faults, from data acquisition to optimization, has been achieved. The data acquisition module, with the help of multi-source sensors, can comprehensively and in real time capture various operating status data of the refrigeration unit, breaking through the limitations of traditional manual inspections, making the perception of equipment status more comprehensive and timely, and avoiding misjudgment or omission of faults due to missing information.

[0057] The feature extraction module extracts data reflecting equipment operating characteristics from massive amounts of operational status data, transforming the chaotic raw data into feature information with practical analytical value, thus providing a precise analytical target for subsequent fault assessment. This in-depth data processing avoids the drawbacks of simply piling up raw data, providing a more targeted basis for fault diagnosis.

[0058] The feature baseline values ​​generated by the fault assessment module provide an objective standard for judging differences in equipment operating status. Based on these baseline values, the degree of difference in operating status is analyzed and initial fault parameters are generated, changing the traditional subjective mode that relies on experience-based judgment. Through quantitative difference analysis, the presence and severity of equipment anomalies can be reflected more objectively, making fault assessment free from interference by human factors and improving the consistency and reliability of fault identification.

[0059] The analysis and optimization module determines the optimization direction based on the initial fault parameters and matches it with historical fault data to generate fault optimization signals. This process fully utilizes the value of historical fault data, transforming past fault handling experience into a reference for current fault resolution, avoiding repetitive work and accelerating the response speed of fault handling. Simultaneously, by reusing historical data, possible solution paths can be quickly found when facing new faults, reducing the randomness of fault troubleshooting.

[0060] The cloud service module transforms fault optimization signals into cloud platform control commands, enabling remote and centralized management of fault handling. Leveraging the advantages of the cloud platform, geographical limitations can be overcome, allowing for unified fault monitoring and handling of refrigeration units in different regions and of different models. Simultaneously, the cloud platform facilitates centralized data storage and sharing, providing convenience for cross-device fault analysis and experience accumulation, and promoting an overall improvement in fault handling capabilities. Attached Figure Description

[0061] Figure 1 This is a timing diagram of the cloud service system for refrigeration unit fault analysis described in this invention;

[0062] Figure 2 A flowchart illustrating the operation of the feature extraction module;

[0063] Figure 3 A flowchart for generating feature baseline values;

[0064] Figure 4 A flowchart generated for initial fault parameters;

[0065] Figure 5 This is a flowchart for calculating feature similarity. Detailed Implementation

[0066] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] Please see Figure 1 The present invention provides a cloud service system for refrigeration unit fault analysis. The system includes a closed-loop processing flow comprising a data acquisition module, a feature extraction module, a fault assessment module, an analysis and optimization module, and a cloud service module.

[0068] The data acquisition module uses vibration sensors, temperature sensors, and pressure sensors installed at key parts of the refrigeration unit to collect real-time vibration signals of the compressor shaft, condenser inlet and outlet temperature data, and evaporator pressure change data, with a sampling frequency set to 10kHz. The feature extraction module uses a digital signal processor to preprocess the raw data, including wavelet noise reduction and moving average filtering. The fault assessment module establishes a feature baseline model based on a deep belief network. The analysis and optimization module is equipped with a distributed fault database storing historical fault cases from the past five years. The cloud service module interacts with the cloud control platform via a RESTful API.

[0069] Example 1: See Figure 2 The data acquisition module is equipped with a multi-source sensor network to achieve three-dimensional monitoring of the refrigeration unit's operating status. Compressor vibration monitoring uses a triaxial piezoelectric accelerometer, installed on the drive end, non-drive end bearing housing, and motor housing, with a sampling frequency of 12.8kHz and a measurement range covering ±50g. The temperature monitoring system arranges eight PT100 platinum resistance thermometers along the condenser tube side, using a three-wire connection to eliminate the influence of wire resistance, achieving a measurement accuracy of ±0.3℃, and acquiring temperature values ​​twice per second to construct the original temperature curve. Evaporation pressure monitoring uses a diffused silicon piezoresistive transmitter, with dual measuring points at the evaporator inlet and outlet. The 4-20mA analog signal is converted into a digital pressure sequence by a 24-bit ADC.

[0070] The feature extraction module executes a hierarchical signal processing flow. In the vibration spectrum processing stage: the original vibration signal is first filtered by an 8th-order Butterworth high-pass filter to remove mechanical interference below 5Hz, and then a 2048-point fast Fourier transform is performed to obtain the spectrum. The fundamental frequency and the peak values ​​of the 2nd to 6th harmonic components are located within the 50-800Hz operating frequency band, and the amplitude values ​​of each order are accurately recorded to generate a set of dominant frequency amplitudes. Temperature sequence processing employs adaptive smoothing technology: a dynamic window width Savitzky-Golay filter is applied to the original temperature curve, automatically adjusting the polynomial order (2nd-4th order) and window size (5-25 points) according to the temperature change rate, eliminating temperature step phenomena caused by the cooling cycle, and forming a stable temperature sequence with fluctuations controlled within ±0.5℃. Pressure change rate calculation establishes a differential model: using a 10-second time unit, the instantaneous slope of pressure change is calculated using the central difference method. When pressure pulse interference is detected, it automatically switches to a weighted moving average algorithm, finally outputting the unit time fluctuation parameter.

[0071] Data from each feature dimension is integrated using a specific data structure. The fundamental frequency amplitude set uses a six-dimensional floating-point array to store the fundamental frequency and harmonic amplitude values, along with a spectrum acquisition time stamp. The stable temperature sequence stores the most recent 300 temperature sampling points in a circular buffer, with a data refresh period of 500 milliseconds. The unit-time fluctuation record includes three indicators: the current value, the mean of the previous 60 seconds, and the standard deviation. Feature encapsulation uses a binary protocol, with the device ID and timestamp set in the header, the data segment storing three sets of feature vectors in TLV format, and a CRC32 checksum at the tail.

[0072] The sensor network establishes dual communication paths. Vibration sensors directly transmit raw waveform data via the PROFINET industrial Ethernet protocol, while the temperature and pressure sensor group uses an RS-485 bus to build a MODBUS-RTU network. Dual-channel data is time-aligned at the edge computing gateway, with time synchronization accuracy controlled within ±10 milliseconds. The data preprocessing process implements a parallel pipeline design: vibration signal processing time is controlled within 15 milliseconds, temperature filtering cycle does not exceed 20 milliseconds, and pressure differential calculation is maintained at the 10-millisecond level, ensuring a complete feature vector update is completed every 30 seconds.

[0073] A dynamic calibration mechanism is configured for feature vectors. The main frequency amplitude set is automatically associated with the current compressor speed, and spectral recalibration is triggered when the speed fluctuation exceeds ±5%. An ambient temperature compensation model is established based on the stable temperature sequence, and the reference value is dynamically corrected according to the condenser inlet air temperature. A condition adaptation algorithm is implemented for the unit time fluctuation, and the reference benchmark is automatically reset when the cooling load change exceeds 30%. The feature database adopts a hierarchical storage strategy: the memory buffer retains the most recent 8 hours of high-frequency feature data, the solid-state drive stores 30 days of historical feature sets, and the hard disk drive enables five-year cold data archiving.

[0074] The system establishes a characteristic quality monitoring system. A signal-to-noise ratio threshold is set for the vibration spectrum; when the background noise exceeds the fundamental frequency amplitude by 20%, a sensor self-test is triggered. The standard deviation of the temperature sequence is calculated in real time; if the standard deviation exceeds the limit after 10 consecutive samplings, a backup sensing channel is switched on. A change rate threshold is configured for pressure fluctuations; if an instantaneous change exceeds 50% of the measurement range, the system automatically switches to manual verification mode. This implementation method achieves accurate conversion of operational status data into multi-dimensional feature vectors, establishing a standardized data input system for subsequent fault analysis.

[0075] Example 2: See Figure 3 The process for generating characteristic baselines and calculating operational status differences is as follows: The characteristic baseline value is constructed by integrating three core indicators: the historical dominant frequency amplitude average reflects the long-term operating characteristics of the compressor; the temperature fluctuation threshold indicates the stability boundary of the condensing system; and the historical pressure fluctuation range depicts the normal operating range of the evaporation process. The historical dominant frequency amplitude average is obtained using a spatiotemporal correlation analysis method. The system automatically matches vibration spectrum data (accurate to the hour) from the same season, the same working day, and the same time period over the past three years to establish a prediction model based on Gaussian process regression. When the historical data sample size is insufficient, the system switches to expanding the search to adjacent time periods. This average calculation incorporates an environmental temperature and humidity compensation coefficient. Under air conditioning conditions, the humidity and heat environment correction parameter Φ (range 0.85-1.15) is enabled, and its calculation formula is as follows:

[0076]

[0077] in: This represents the average amplitude of the dominant frequency of the vibration after compensation. It is the amplitude measurement value of the kth historical data point. The environmental correction parameters are for the corresponding time period, and n is the number of valid historical data points.

[0078] Temperature fluctuation threshold settings must differentiate between system operating modes. In cooling mode, a correlation matrix is ​​established between condensing temperature and cooling water flow rate; for every 0.5 m / s increase in cooling water flow rate, the threshold is relaxed by 0.3℃. In heating mode, the threshold coefficient is dynamically adjusted based on the ambient wet-bulb temperature. This threshold employs a two-layer control logic: the base threshold is determined by the equipment nameplate parameters, and the dynamic adjustment layer automatically adjusts the range by ±0.5℃ based on recent operational stability indices. Historical pressure fluctuation interval calculations utilize a rolling time window statistical method, updating the reference dataset every 6 hours and removing data points from start-up and shutdown transition periods. The system establishes a pressure-load correlation model; when the cooling load change exceeds 15%, an interval recalculation is immediately triggered.

[0079] Feature normalization preprocessing is implemented during weighted fusion. Vibration amplitude data is converted into percentage values ​​by dividing by the equipment's rated amplitude value, temperature fluctuation thresholds are mapped to relative coefficients in the [0,1] interval, and pressure fluctuations are standardized using Z-score. A dynamic adjustment mechanism for fusion weight configuration is implemented: fixed weights are used during the initial system startup phase (first 72 hours of operation); during the stable operation phase, a learning-based weight adjuster is activated to periodically optimize weight allocation based on the historical fault detection rate of each feature dimension.

[0080] The total dissimilarity score is constructed using a non-linear transformation model. The weighted summation of the individual scores is then input into a sigmoid function for standardization to a percentage level.

[0081]

[0082] in: The final difference score, Weighting coefficients corresponding to vibration, temperature, and pressure characteristics (initial values ​​0.5, 0.3, and 0.2). The function assigns scores to each sub-item. In the low-band (D<30), a gentle curve is used to avoid false alarms, while in the high-band (D>70), a steep slope is employed to enhance fault sensitivity.

[0083] Real-time performance is ensured throughout the calculation process. Vibration spectrum analysis utilizes an embedded DSP core for millisecond-level processing, while temperature duration detection incorporates a hard timer module to avoid software timing errors. Pressure interpretation employs a dedicated comparator circuit for microsecond-level response. A continuous monitoring mechanism is implemented for the calculation results: when the difference exceeds a critical threshold (40 points) for three consecutive samples, a verification process is initiated, recalculating and verifying the results using redundant sensor data. The dynamic baseline update module is configured with a conflict detection mechanism, automatically freezing baseline updates during periods of sudden system load changes to prevent feature drift.

[0084] This implementation method establishes the ability to capture abnormal evolution trends. A double exponential smoothing prediction model is used for amplitude shifts, predicting amplitude trends three periods in advance; a second-order differential monitor is set up for temperature exceedances, automatically raising the anomaly level when the exceedance rate accelerates; directional trend discrimination is implemented for pressure fluctuations, triggering a pre-alarm mechanism after five consecutive exceedances in the same direction. Historical data compression and storage employs an improved rotating door algorithm, reducing storage consumption by 85% while retaining key features such as lower limit points and peak points, supporting online storage of five years of original data.

[0085] Example 3: See Figure 4 This paper details the mechanism for determining the difference level threshold and matching it with historical fault data. The difference level threshold employs a dynamic interval division method, establishing a three-level automatic threshold adjustment model based on the equipment's cumulative operating hours. The first difference level threshold... The score changes in a stepped manner with operating time: 65 points for 0-2000 hours of equipment operation, 70 points for 2001-8000 hours, and 75 points for over 8001 hours. Second difference level threshold. Always more The upper 15-point fluctuation range, the third difference level threshold The score is kept constant at 95. The threshold comparator is implemented using hardware logic circuitry, with a response time controlled within 50 nanoseconds.

[0086] The initial fault parameter generation module includes three types of parallel processing channels. The compressor frequency regulation parameter calculation employs a variable step-size search algorithm. The refrigerant flow correction parameter is generated by a fuzzy PID controller, with inputs being the evaporator pressure fluctuation rate ε and the temperature over-limit ratio η, and the output being the expansion valve opening adjustment coefficient κ∈[-0.3, 0.3]. The system shutdown detection parameters include eight self-test item codes, ordered by priority as follows: oil pressure status code... Winding insulation code Refrigerant purity code Each test item has an independent timeout threshold. Second( (Priority number).

[0087] The historical fault data set employs a spatiotemporal four-dimensional index structure. The first dimension categorizes faults by device model, establishing a device fingerprint database containing 32 feature parameters. The second dimension classifies faults by type, using an improved LOF algorithm to automatically cluster similar faults. The third dimension records the occurrence time, accurate to millisecond timestamps. The fourth dimension stores environmental condition snapshots, including 12 environmental parameters such as dry / wet bulb temperature and altitude. When retrieving data, composite query conditions are constructed, with the default search radius set to a feature deviation range of ±15% and a time window of ±2 hours.

[0088] The feature matching process employs a multi-stage screening strategy. The initial screening uses a Bloom filter to quickly eliminate irrelevant records, with a false positive rate tolerance of 0.1%. The secondary matching applies an improved Jaccard similarity calculation, discretizing continuous features using equal-width binning. The precise comparison stage employs a dynamic time warping algorithm, allowing for ±5% time axis scaling.

[0089] Multi-objective historical feature processing employs a conflict resolution mechanism. When parameter directional conflicts occur, a DS synthesis rule based on evidence theory is initiated to calculate the confidence interval. For compressor frequency parameters, a maximum adjustment range constraint is set to prevent directional reversal between adjacent adjustment cycles. Refrigerant flow correction implements a gradual adjustment strategy, decomposing historical recommended values ​​into three step-like change stages, with each stage spaced at least 5 minutes apart.

[0090] The system establishes a case backtracking and verification mechanism. After each parameter optimization implementation, the feature change trend is continuously monitored over the next 30 minutes, comparing the actual improvement effect with the expected target. Verification results are categorized into three types: complete match, partial match, and no match, used to dynamically adjust the confidence weight of the case in the historical database. For newly emerging unmatched feature combinations, an entry for a pending fault case is automatically generated, requested to be annotated by human experts, and then added to the knowledge base. Data storage adopts a columnar compression format, with the average space occupied by a single fault case controlled within 2KB, supporting more than 1000 concurrent queries per second.

[0091] Example 4: See Figure 5 This document details the process of feature similarity calculation and fault optimization signal generation. After receiving the feature baseline values ​​from the fault assessment module, the system initiates a multi-level feature matching process. Taking a certain model of centrifugal chiller unit as an example, the current feature baseline values ​​include three core parameters: vibration dominant frequency amplitude of 42μm, condensing temperature fluctuation range of ±1.2℃, and evaporation pressure change rate of 0.8kPa / s. The historical fault database stores 876 fault records of similar equipment from the past three years, each record containing a 12-dimensional feature vector and corresponding optimization measures.

[0092] The characteristic distance calculation employs a hybrid metric system. Vibration spectrum distance is obtained through piecewise integration, dividing the 20-500Hz frequency band into eight characteristic sub-bands and calculating the sum of the absolute values ​​of the energy differences between each sub-band. Temperature fluctuation mode distance incorporates a dynamic time warping algorithm, aligning temperature fluctuation curves at different time scales to calculate path deviation. Pressure change rate distance uses an improved editing distance algorithm, considering the consistency of the trend direction. The system sets an adaptive distance threshold, automatically adjusting the matching sensitivity based on the current operating load rate: the threshold is relaxed by 15% when the load rate is 70%-100%, and tightened by 20% when the load rate is below 50%.

[0093] A dual verification mechanism is implemented for the screening of valid historical data. The primary screening retains records whose feature distance is less than 1.5 times the equipment type benchmark value, while the secondary screening requires that at least two auxiliary features (such as oil temperature and current harmonics) have a matching degree of more than 60%. The intermediate results of a certain matching process are shown in Table 1.

[0094] Table 1: Intermediate results of a matching process.

[0095] Fault ID Vibration distance Temperature distance Pressure distance Overall matching degree Historical optimization measures F2019-028 0.32 0.45 0.21 72% Frequency decreased by 5% + refrigerant increased by 8% F2020-153 0.28 0.51 0.33 68% Frequency drop of 3%+ condenser cleaning F2021-077 0.41 0.38 0.25 75% Replace the oil filter + frequency reduced by 7% F2022-214 0.35 0.42 0.29 70% Increase refrigerant by 5% + adjustment of expansion valve

[0096] The quantity percentage statistics employ a sliding window analysis. The system automatically divides the data into six time-dimensional windows: the most recent 7 days, historical data for the same season, historical data for the same load period, historical data for the same ambient temperature period, historical data for the same maintenance cycle, and the entire historical range. The percentage of valid cases is calculated independently for each window, and the final weight is obtained by summing the product of the window importance coefficient (0.2-0.35) and the percentage value. When the percentage in a certain window falls below 5%, an expert review process is automatically triggered to prevent random matching from dominating the decision-making process.

[0097] The fault optimization signal generation module includes three types of instruction builders. The first optimization instruction signal directly uses historical optimization records, but adds a parameter verification step: checking whether the suggested frequency adjustment exceeds the current equipment's allowable range and whether refrigerant adjustments conflict with recent maintenance records. The second optimization instruction signal performs parameter fusion calculations, taking the weighted median of suggested values ​​from multiple matching cases to eliminate the influence of extreme values. The third optimization instruction signal implements a gradual adjustment strategy, decomposing the total optimization amount into 3-5 tiered steps, with each step spaced within an equipment response monitoring period.

[0098] The signal encoding adopts a layered protocol structure. The base layer contains device identification codes, timestamps, and instruction type markers; the parameter layer stores adjustment values ​​categorized by optimization type, with frequency parameters accurate to 0.1Hz and flow parameters retained to two decimal places; the control layer specifies execution condition constraints, including metadata such as ambient temperature range and maximum execution duration. The transmission protocol is configured with a forward error correction mechanism, adding Reed-Solomon error correction codes to each frame of data to ensure complete instruction recovery even with a 10% packet loss rate.

[0099] The system implements a closed-loop optimization and verification mechanism. After each instruction is executed, the feature monitoring module continuously tracks the slope of change in six key indicators. When the actual improvement rate is lower than 60% of the expected value, some parameter adjustments are automatically rolled back and a second matching is initiated. For successfully optimized cases, the system records the feature response pattern and updates the confidence score of the case library. New cases must pass similarity detection before being added to the library to prevent excessive accumulation of duplicate cases. Data storage adopts column-family partitioning, storing frequently accessed recent cases in an SSD storage pool, while historical cases are archived using a columnar format with a compression ratio of 15:1.

[0100] The case maintenance module implements lifecycle management. Each fault record is set to have a validity period of 36 months; expired cases are automatically transferred to a pending verification state. The system performs a monthly case quality assessment, marking cases that fail verification three times consecutively as invalid. Maintenance engineers can adjust case weights using a visual tool, but all modifications must undergo two-factor authentication and be logged in the audit log. A knowledge base synchronization mechanism ensures case consistency among distributed nodes, and a vector clock algorithm is used to resolve version conflicts.

[0101] Example 5: Focusing on the generation and execution mechanism of cloud platform control commands. After receiving the fault optimization signal, the cloud service module initiates the command compilation process, with three types of input signals corresponding to differentiated command generation logic. When the first optimization command signal is detected, the command compiler parses the specific parameters in the historical optimization record, extracting core elements such as the compressor target frequency value, adjustment duration, and transition curve type. The target frequency value needs to be verified through the equipment capability model, automatically eliminating invalid parameters that exceed physical limits. The adjustment curve generation adopts a seven-segment S-shaped acceleration and deceleration algorithm, with the initial acceleration controlled within 0.3 m / s² to avoid mechanical shock. An environmental compensation factor is added to the command parameter package, automatically reducing the maximum adjustment range by 20% when the computer room temperature exceeds 35℃.

[0102] A multi-parameter coordination engine is introduced to process the second optimization command signal. The refrigerant flow stage adjustment command is divided into three execution cycles, each lasting a fixed 120 seconds. The first cycle performs basic flow correction, with the adjustment range not exceeding 40% of the total recommended value; the second cycle implements dynamic correction, fine-tuning the flow coefficient based on the evaporation pressure feedback value at the end of the first cycle; the final cycle performs steady-state optimization, gradually converging to the target flow value. A flow change rate constraint valve is set for each stage, prohibiting a single-cycle flow change exceeding 3% of the system's total rated flow. A safety monitoring protocol is embedded in the command package, automatically pausing stage progression when a reverse fluctuation of more than 10% is detected between two adjacent pressure sampling points.

[0103] The third optimization command signal triggers a system-level diagnostic response. The retest command includes a four-level detection sequence: the oil pressure system test (priority 0) is completed within 150 milliseconds, collecting lubricating oil pressure, oil temperature, and particulate matter concentration; the refrigerant purity analysis (priority 1) lasts 35 seconds, detecting the refrigerant composition ratio using a spectral sensor; the electrical insulation test requires disconnecting the power supply and takes 18 seconds to complete five tests, including winding insulation resistance and grounding continuity; the auxiliary system test covers peripheral equipment such as water pump status and cooling tower fans. The test items implement a parallel start strategy, with non-conflicting test items activated simultaneously, and the overall retest process is completed within 300 seconds.

[0104] Command transmission establishes a triple-protection channel. The primary transmission uses the MQTT protocol, with a QoS level of 1 and a "at least once" guarantee mechanism. Topic naming follows a tree structure of device geographic partition / device model / control type. The secondary channel sends command digests via LoRaWAN, containing the command type code and core parameter checksum. The backup channel uses an SMS gateway to transmit Base64-encoded simplified commands, with the character length compressed to within 160 bytes. Real-time link quality monitoring is implemented during transmission, automatically switching to the secondary channel when the MQTT latency exceeds 500 milliseconds. Each command packet carries a triple checksum sequence: a CRC-16 cyclic redundancy check in the header, a command feature value hash check in the middle, and a timestamp dynamic key verification at the tail.

[0105] The command execution unit is configured with a response monitoring system. The compressor actuator interface uses 4-20mA analog signal control, with an additional digital status feedback pin. The refrigerant regulating valve is equipped with a dual-loop position sensor to compare the valve position command with the actual opening deviation in real time. The system retest unit is equipped with an independent watchdog timer, which automatically sends an interrupt pulse when a certain test timeout occurs. All execution processes are implemented with closed-loop data recording; the baseline parameters before execution, the instantaneous status during execution, and the stable values ​​after execution are all uploaded to the cloud platform log system.

[0106] The fault switching mechanism establishes a multi-level response strategy. For compressor control commands, if a sudden increase in shaft vibration exceeding 25% is detected during execution, the adjustment is immediately interrupted and the original frequency is restored. If an evaporator temperature becomes too low (below the set value by 5°C for 20 seconds) during refrigerant regulation, a reverse correction is automatically executed to restore the system to its pre-adjustment state. Severe faults detected during system retesting (such as insulation resistance below 2MΩ) directly trigger the system shutdown sequence, simultaneously starting the standby unit. Standby unit switching involves five interlocking actions: closing the faulty unit's refrigerant main valve → starting the standby unit's pre-lubrication pump → standby unit's inverter soft start → refrigerant bypass valve switching → load transfer control. The total switching time is limited to within 180 seconds, and the load transfer rate is controlled to increase by 10% of the rated load per hour.

[0107] The historical command database is managed using versioning. Each generated command package is assigned a unique version number in the format of Device ID-Timestamp-Command Type. Execution results are archived in three states: fully executed, partially executed, and unexecuted. Maintenance personnel can simulate system responses using the historical command replay function, but the actual control port is locked during replay. The system automatically generates a monthly command execution performance report, including key indicators such as response latency distribution, execution achievement rate, and fault recovery time. The command template library is updated based on quarterly evaluation results, eliminating outdated templates with a usage rate of less than 5% for six consecutive times.

[0108] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0109] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cloud service system for refrigeration unit fault analysis, characterized in that, include: Data acquisition module: Acquires real-time operating status data of the refrigeration unit through multi-source sensors; Feature extraction module: Receives the operating status data and extracts the operating characteristic data of the chiller unit; Fault assessment module: Generates characteristic baseline values ​​based on the operating characteristic data of the refrigeration unit; The degree of difference in operating status is determined based on the characteristic baseline value, and initial fault parameters are generated based on the degree of difference in operating status. Analysis and optimization module: Determines the parameter optimization direction based on the initial fault parameters; acquires a historical fault data set based on the parameter optimization direction; matches the feature baseline value with the historical fault data set, and generates a fault optimization signal based on the matching result; Cloud service module: Receives the fault optimization signal and generates cloud platform control commands; The process of receiving the operating status data and extracting the operating characteristic data of the chiller unit includes: Obtain the compressor vibration spectrum, condensing temperature time-series curve, and evaporating pressure change rate from the operating status data; The vibration spectrum of the compressor is decomposed in the frequency domain to obtain the set of main frequency amplitudes; The condensation temperature time series curve is smoothed to obtain a stable temperature sequence. Calculate the unit-time fluctuation of the evaporation pressure change rate; The main frequency amplitude set, stable temperature sequence and unit time fluctuation are integrated into the operating characteristic data of the refrigeration unit; The step of generating characteristic baseline values ​​based on the operating characteristic data of the refrigeration unit includes: Extract the historical average amplitude of the main frequency during the same period from the operating characteristic data of the refrigeration unit; Obtain the fluctuation threshold of the stable temperature sequence; Calculate the historical normal range of the fluctuation per unit time; The historical average amplitude of the dominant frequency, the fluctuation threshold, and the historical normal interval are weighted and fused to generate a characteristic baseline value; The determination of operational status differences based on the feature baseline value includes: The amplitude offset is obtained by calculating the difference between the current dominant frequency amplitude and the average dominant frequency amplitude of the same period in history; The duration for which the stable temperature sequence exceeds the fluctuation threshold is detected; Determine whether the fluctuation per unit time deviates from the historical normal range; The operational state difference is generated based on the amplitude offset, duration, and disengagement state. The step of generating initial fault parameters based on the operational state difference includes: The difference in operating status is compared with a preset difference level threshold; When the difference in operating status reaches the first difference level threshold, compressor frequency adjustment parameters are generated as initial fault parameters. When the difference in operating status reaches the second difference level threshold, refrigerant flow correction parameters are generated as initial fault parameters. When the difference in operating status reaches the third difference level threshold, system shutdown detection parameters are generated as initial fault parameters.

2. The cloud service system for refrigeration unit fault analysis according to claim 1, characterized in that, The step of matching the feature baseline value with the historical fault data set includes: Search the historical fault data set for a target historical feature that is the same as the feature baseline value; When a single target historical feature exists, retrieve the historical optimization record corresponding to that target historical feature; When multiple target historical features exist, calculate the average optimization coefficient of the multiple target historical features; When no target historical features exist, the feature similarity calculation process is initiated.

3. The cloud service system for refrigeration unit fault analysis according to claim 2, characterized in that, The startup feature similarity calculation process includes: Calculate the characteristic distance between the characteristic baseline value and each historical fault data; Filter valid historical data where the feature distance is less than the distance threshold; The percentage of valid historical data is calculated. The feature matching weight is determined based on the stated proportion of quantities.

4. The cloud service system for refrigeration unit fault analysis according to claim 3, characterized in that, The step of generating a fault optimization signal based on the matching result includes: When historical optimization records are obtained, the historical optimization records are used to generate the first optimization instruction signal; When the average optimization coefficient is obtained, the average optimization coefficient is fused with the current initial fault parameters to generate a second optimization command signal; When the feature matching weight is obtained, the initial fault parameters are adjusted according to the feature matching weight to generate a third optimization instruction signal; The fault optimization signal includes a first optimization command signal, a second optimization command signal, or a third optimization command signal.

5. The cloud service system for refrigeration unit fault analysis according to claim 4, characterized in that, The step of receiving the fault optimization signal and generating cloud platform control commands includes: Upon receiving the first optimization command signal, a compressor frequency recalibration command is generated; When the second optimization command signal is received, a refrigerant flow stage adjustment command is generated. When the third optimization instruction signal is received, a system retest instruction is generated; The cloud platform control commands drive the refrigeration unit to perform corresponding operations.

6. A cloud service control method for refrigeration unit fault analysis, characterized in that, It includes all modules and method flows of the cloud service system for refrigeration unit fault analysis as described in any one of claims 1 to 5.

Images