A system for analyzing the operating state of hydraulic engineering equipment

Abstract

The application relates to the technical field of predictive maintenance and health management of water conservancy engineering equipment, in particular to a running state analysis system for water conservancy engineering equipment, which comprises the following: a data acquisition unit that acquires a data set in real time; a stress baseline calculation unit that calculates data to obtain a hydraulic stress baseline; an intrinsic residual analysis unit that performs vector difference calculation to obtain a physical calibration residual vector and a scalar amplitude; a preset statistical fault triggering threshold is discriminated and processed, a working condition stress signal or an intrinsic fault signal is obtained; a fault attribution unit updates a current health state vector of the equipment in combination with the scalar amplitude and the duration; a scheduling avoidance unit calls a digital twin reference model, predicts an operation consequence and calculates a weighted risk index, when the weighted risk index exceeds a preset risk threshold, a planned operation instruction is refused to be executed; the application improves the sensitivity and accuracy of early fault identification, and effectively avoids frequent false alarms caused by transient working condition impact.

Description

Technical Field

[0001] This invention relates to the field of predictive maintenance and health management technology for water conservancy engineering equipment, specifically to an operational status analysis system for water conservancy engineering equipment. Background Technology

[0002] As the structure and operating environment of large-scale water conservancy equipment become increasingly complex, they are subjected to drastic and frequent changes in dynamic hydraulic conditions during operation, which results in extremely complex condition monitoring data for the equipment, making it a hybrid entity.

[0003] Currently, traditional equipment condition monitoring methods typically rely on setting fixed thresholds for sensor signals for evaluation. The raw monitoring data collected is a mixture of the equipment's own health status and external hydraulic conditions. When hydraulic conditions change drastically, even if the equipment is perfectly healthy, its sensor readings may fluctuate wildly, making it difficult to distinguish from actual fault signals. Therefore, this method cannot effectively isolate strong interference from external operating conditions. True early, minor fault signals are often drowned out by strong background noise from operating conditions, resulting in insufficient sensitivity and accuracy in fault detection and a large number of false alarms. Therefore, how to accurately separate the intrinsic fault signals that truly reflect the physical degradation of the equipment from the mixed signals of strong operating condition interference, and thus achieve accurate assessment of the equipment's health status and early fault warning, is a technical problem that needs to be solved in the field of water conservancy engineering. Summary of the Invention

[0004] To solve the above-mentioned technical problems, the present invention provides an operational status analysis system for water conservancy engineering equipment. Specifically, the technical solution of the present invention includes:

[0005] The data acquisition unit is used to acquire global hydraulic condition vectors and raw mixed state datasets in real time;

[0006] The stress baseline calculation unit is used to perform physical simulation calculations on the global hydraulic condition vectors collected by the data acquisition unit based on the preset digital twin reference model and the preset absolute health state, so as to obtain the hydraulic stress baseline.

[0007] The intrinsic residual analysis unit is used to perform vector difference calculation on the original mixed state dataset and the hydraulic stress baseline to obtain the physical calibration residual vector and scalar amplitude. It also performs discrimination processing on the scalar amplitude and the preset statistical fault trigger threshold to obtain the working condition stress signal or intrinsic fault signal.

[0008] When an intrinsic fault signal is generated, the fault attribution unit is used to input the physical calibration residual vector at the time of triggering the alarm into a pre-trained fault mode classifier to identify the fault type, and combine the scalar amplitude and duration to calculate the cumulative damage offset through a preset damage accumulation model, thereby updating the current health status vector of the device.

[0009] The scheduling avoidance unit is used to respond to the received planned operation instructions, call the digital twin reference model, and input the global hydraulic condition vector collected in real time by the data acquisition unit, the planned operation instructions, and the current health status vector of the equipment updated or maintained by the fault attribution unit. It predicts the operation consequences and calculates the weighted risk index; and compares and analyzes the weighted risk index with the preset risk threshold. When the weighted risk index exceeds the preset risk threshold, the planned operation instructions are rejected.

[0010] Preferably, the statistical fault triggering threshold is defined as follows: calculate the statistical distribution of the scalar amplitude within a known healthy operating cycle to obtain the mean and standard deviation; multiply the mean and standard deviation by a preset dimensionless sensitivity coefficient and add them together to obtain the statistical fault triggering threshold.

[0011] Preferably, the global hydraulic condition vector includes upstream peak flow, downstream characteristic water level, gate opening sequence, and pump speed; the original mixed state dataset includes vibration signal vector, temperature of key components, and inlet and outlet pressure.

[0012] Preferably, the process by which the fault attribution unit updates the current health status vector of the device is as follows: mapping the output of the fault mode classifier to specific physical parameters in the digital twin reference model; and simultaneously, inputting the scalar amplitude and duration into a preset damage accumulation model to quantitatively calculate the cumulative damage offset of the specific physical parameters.

[0013] Preferably, the baseline vector corresponding to the preset absolute health state is combined with all identified cumulative damage offset vectors to dynamically maintain and output the current health state vector of the device.

[0014] Preferably, the calculation process of the weighted risk index is as follows: traverse all key sensor channels; for each channel, obtain the model's predicted output value for that sensor, the preset safety alarm threshold, and the physical limit threshold; calculate the portion of the predicted output value that exceeds the safety alarm threshold, and divide it by the difference between the physical limit threshold and the safety alarm threshold to obtain the normalized degree of exceeding the limit.

[0015] Preferably, the normalized exceedance degree of each channel is multiplied by the corresponding preset dimensionless risk weight, and the weighted results of all channels are summed to obtain the dimensionless weighted risk index.

[0016] Preferably, the scalar magnitude is obtained by acquiring the L2 norm or Euclidean distance of the physical calibration residual vector.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] 1. Using a digital twin model, based on the real-time collected global hydraulic condition vector, a hydraulic stress baseline that purely reflects the impact of the operating conditions is physically simulated and calculated. By performing vector difference between the actual collected raw mixed state dataset and this baseline, the interference caused by drastic changes in operating conditions such as upstream flood peaks and downstream water levels is effectively eliminated. This successfully decouples and separates weak intrinsic fault signals of equipment from strong background noise, significantly improving the sensitivity and accuracy of early fault identification.

[0019] 2. Traditional fixed thresholds are prone to false alarms triggered by drastic fluctuations in operating conditions. This solution compares the scalar amplitude of the physical calibration residual vector with the dynamic threshold. This statistical fault trigger threshold is adaptively generated based on the statistical distribution of the scalar amplitude when the equipment responds to various operating conditions within a known health cycle. This makes the threshold scientifically include the normal fluctuation range, which can distinguish real physical degradation with extremely high specificity and effectively avoid frequent false alarms caused by transient operating condition shocks.

[0020] 3. When an intrinsic fault signal is generated, the fault attribution unit no longer only outputs an abstract alarm. It uses a pre-trained fault mode classifier to identify the fault type and maps it to a specific physical parameter in the digital twin model. At the same time, it combines the scalar amplitude and duration of the residual and quantitatively calculates the cumulative damage offset of the physical parameter through the damage accumulation model. By combining the baseline state and the damage offset, it dynamically updates the equipment health status vector, realizing accurate assessment from abstract alarm to physical parameter quantification.

[0021] 4. This solution achieves a shift from passive response to proactive avoidance. After receiving the planned operation instruction, the scheduling and avoidance unit invokes the digital twin model and inputs the real-time operating conditions, operation instructions, and the current equipment health status vector with quantified damage to perform a forward-looking risk simulation. By calculating the predicted exceedance degree of each channel and weighting the sum, a dimensionless weighted risk index is obtained. When this index exceeds a preset threshold, the system will proactively refuse to execute the instruction, thereby preventing dangerous operations that would occur in the current degraded state of the equipment. Attached Figure Description

[0022] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0023] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0025] Example 1

[0026] Please see Figure 1 A system for analyzing the operational status of water conservancy engineering equipment, comprising:

[0027] The data acquisition unit is used to acquire global hydraulic condition vectors and raw mixed state datasets in real time;

[0028] The stress baseline calculation unit is used to perform physical simulation calculations on the global hydraulic condition vectors collected by the data acquisition unit based on the preset digital twin reference model and the preset absolute health state, so as to obtain the hydraulic stress baseline.

[0029] The intrinsic residual analysis unit is used to perform vector difference calculation on the original mixed state dataset and the hydraulic stress baseline to obtain the physical calibration residual vector and scalar amplitude. It also performs discrimination processing on the scalar amplitude and the preset statistical fault trigger threshold to obtain the working condition stress signal or intrinsic fault signal.

[0030] When an intrinsic fault signal is generated, the fault attribution unit is used to input the physical calibration residual vector at the time of triggering the alarm into a pre-trained fault mode classifier to identify the fault type, and combine the scalar amplitude and duration to calculate the cumulative damage offset through a preset damage accumulation model, thereby updating the current health status vector of the device.

[0031] The scheduling avoidance unit is used to respond to the received planned operation instructions, call the digital twin reference model, and input the global hydraulic condition vector collected in real time by the data acquisition unit, the planned operation instructions, and the current health status vector of the equipment updated or maintained by the fault attribution unit. It predicts the operation consequences and calculates the weighted risk index; and compares and analyzes the weighted risk index with the preset risk threshold. When the weighted risk index exceeds the preset risk threshold, the planned operation instructions are rejected.

[0032] This embodiment provides an operational status analysis system for water conservancy engineering equipment, characterized in that it comprises five core units, which together form a complete technical closed loop from data acquisition, status decoupling, fault attribution to proactive avoidance;

[0033] The data acquisition unit aims to provide real-time and complete boundary condition and state characterization data for subsequent stress calculations and residual analysis. This unit acquires two key datasets in real time through a sensor array and data interface:

[0034] Global hydraulic condition vector This is a set of parameters defining the hydraulic transient boundary conditions of the system, which originate from external environmental monitoring and equipment control signals; in this embodiment, the vector preferably includes: upstream peak flow. This reflects the main driving source of the input energy; downstream characteristic water level The back pressure condition of the system is defined; the gate opening sequence is defined. and pump unit speed This represents a manual control command;

[0035] Original mixed-state dataset This is a set of sensor measured data characterizing the local operational response of equipment components; in this embodiment, the dataset preferably includes: vibration signal vectors. Temperature of key components and import / export pressure ;

[0036] To clarify the technical premise of this invention, It is mixed data because it reflects both operating conditions and other factors. Instantaneous impact and the health status of the equipment itself The combined effect of these factors can be theoretically defined as follows: ,in It is random noise; the core task of this invention is decoupling. and right Contributions;

[0037] The core purpose of the stress baseline calculation unit is to use a digital twin model to reproduce the sensor readings that an absolutely healthy device should exhibit under current real operating conditions, i.e., the hydraulic stress baseline.

[0038] In this embodiment, the unit first calibrates and solidifies a preset digital twin reference model based on the equipment's factory acceptance data or historical best operating data. The device state corresponding to this model is defined as a preset absolute health state. ;

[0039] This unit will collect the global hydraulic condition vector in real time from the data acquisition unit. As boundary conditions, they are input into the reference model. The physical simulation calculations are performed, and the calculation process can be expressed as follows:

[0040]

[0041] in: That is, the hydraulic stress baseline output by the calculation; this baseline In terms of physical dimensions and vector structure, it is similar to the original mixed-state dataset. Completely consistent; its technical significance lies in the fact that it provides a theoretical reference value for the system response caused purely by changes in operating conditions;

[0042] The intrinsic residual analysis unit aims to isolate the effects of operating conditions and extract the residuals solely caused by the degradation of the equipment's intrinsic condition by comparing actual observed values ​​with theoretical health values. right The difference signal caused by the deviation;

[0043] In this embodiment, the unit receives data from the data acquisition unit. and from the stress baseline calculation unit Performing a rigorous vector difference calculation on the two yields:

[0044] Physical calibration residual vector

[0045] because and Sharing the same operating condition input The difference The working condition was theoretically eliminated. The impact was only retained due to health status. Deviation The resulting differences;

[0046] To quantize the total magnitude of the residual vector and eliminate the influence of dimensions among its components, the system first normalizes the residual vector, and then obtains a dimensionless scalar value by acquiring the L2 norm of the normalized vector, which is defined as the scalar magnitude. :

[0047]

[0048] in, This is the normalized physical calibration residual vector;

[0049] This unit has the scalar magnitude With a preset statistical fault trigger threshold Perform real-time discrimination processing; if The system determines that the current sensor fluctuations are within the health statistics range and belong to normal operating condition stress signals, i.e., minor errors between the model and reality caused by drastic changes in operating conditions; if The system then determines that a statistically significant deviation has occurred, triggers the intrinsic fault signal, and activates the fault attribution unit;

[0050] The purpose of the fault attribution unit is to automatically identify the specific type of fault when the intrinsic fault signal is triggered, quantify the cumulative damage caused by the fault, and then dynamically update the current true health profile of the equipment.

[0051] The normal operation of this unit relies on two key models built offline: a pre-trained fault mode classifier, which is trained through supervised learning using historical fault data or simulation data, with the input being typical physical calibration residual vectors generated under different fault modes. The output is a specific fault type label; the preset damage accumulation model: for each known fault type, a corresponding damage accumulation model is established or selected according to its physical degradation mechanism, such as the fatigue crack propagation model based on Paris's law, and the model parameters are calibrated using material property data and experimental data. When the system is running, this unit will directly call these trained and calibrated models.

[0052] In this embodiment, when an intrinsic fault signal is generated, i.e. When triggered, the unit performs the following cooperative operation:

[0053] The fault qualitative identification unit will trigger the physical calibration residual vector at the time of the alarm. Note: A vector with direction and component information is used here. nonscalar The data is input into a pre-trained fault mode classifier; this classifier identifies specific patterns in the residual vector in the multi-dimensional sensor space and outputs specific fault types, such as bearing wear and local cavitation.

[0054] Damage Quantitative Accumulation: The output of the above classifier is used to map to the digital twin reference model. One or more specific physical parameters For example, mapping bearing wear to the friction coefficient parameter in the model; simultaneously, the system incorporates scalar amplitudes. With duration This is then input into a pre-defined damage accumulation model to quantitatively calculate the specific physical parameter. Cumulative damage offset ;

[0055] The health status update system will preset the baseline state vector corresponding to the absolute health status. Compared with all identified cumulative damage offset vectors Combining, for example, through vector addition. This allows for the dynamic maintenance and output of the device's current health status vector. In this embodiment, a linear combination method is used as an example, such as... , where the cumulative damage offset vector Each component can be positive or negative to uniformly characterize the enhancement or attenuation of a specific physical parameter. In more complex application scenarios that require consideration of multi-damage coupling effects, a nonlinear coupling function can also be used. To more accurately describe the interactions between different types of damage;

[0056] The scheduling avoidance unit aims to use an updated digital twin model that reflects the actual degradation state to perform forward-looking simulations of future planned operations in order to prevent high-risk operations that may cause equipment damage or failure.

[0057] In this embodiment, the unit is used to respond to the received planned operation command. For example, a command to quickly raise a pump or open a gate; it does not execute the command immediately, but instead performs the following risk assessment process:

[0058] Calling the digital twin reference model ;

[0059] Input three key vectors: a global hydraulic condition vector acquired in real time by the data acquisition unit. This represents the current operating condition; the received planned operation instructions. ; and crucially, the device's current health status vector, updated or maintained by the fault attribution unit. This represents the actual state of the equipment that has accumulated damage;

[0060] Simulation prediction: Model execution Predict the instantaneous consequences for all critical sensors within a very short time window after the operation is performed;

[0061] Risk quantification: The system calculates a weighted risk index. ;

[0062] Decision: Calculate the weighted risk index With preset risk threshold Perform comparative analysis; when Exceed If the system determines that the operation would pose an unacceptable risk given the equipment's current degraded health condition, it will refuse to execute the planned operation instruction or provide the operator with optimization suggestions.

[0063] The system described in this embodiment constructs a complete analytical framework that integrates data-driven and physical model integration. Its core technical effects are as follows: through stress baseline calculation and intrinsic residual analysis, it achieves for the first time the accurate decoupling and separation of the actual intrinsic fault signals of the equipment from the extremely complex hydraulic stress background noise, greatly improving the detection sensitivity and accuracy of early minor faults; through fault attribution and damage accumulation, it transforms abstract alarm signals into specific physical parameter offsets in the digital twin model, realizing dynamic and accurate quantitative assessment of the equipment's health status; through the scheduling avoidance unit, it uses the updated health model for forward-looking simulation, realizing proactive avoidance of high-risk operations, ensuring the operational safety of hydraulic engineering equipment in complex conditions and continuous degradation processes.

[0064] Example 2

[0065] The statistical fault trigger threshold is defined as follows: Calculate the statistical distribution of the scalar amplitude within a known healthy operating cycle to obtain the mean and standard deviation; multiply the mean and standard deviation by a preset dimensionless sensitivity coefficient and add them together to obtain the statistical fault trigger threshold.

[0066] This embodiment specifically defines the statistical fault triggering threshold used by the intrinsic residual analysis unit. The definition method aims to set a scientific, statistically significant alarm threshold that can dynamically adapt to the specific characteristics of the equipment, so as to distinguish between normal operating stress fluctuations and real intrinsic fault signals.

[0067] In this embodiment, the threshold is defined as follows:

[0068] Data acquisition: Select a segment of data on the operation of the equipment under known health conditions. This data period must include a sufficiently rich range of transient operating conditions, such as start-up, shutdown, speed adjustment, and operating condition switching.

[0069] Residual calculation: Calculate the scalar amplitude at all times within the healthy operating cycle. sequence;

[0070] Statistical distribution calculation: for this health state The sequence was subjected to statistical distribution analysis, and its statistical mean was calculated. and standard deviation ; here This represents the average residual level of the health equipment under model fitting, while This represents the normal fluctuation range of residuals when the health equipment responds to various transient operating conditions;

[0071] Threshold generation: This means... with standard deviation Multiply by the preset dimensionless sensitivity coefficient The statistical fault triggering threshold is obtained by adding the products together. The calculation formula is as follows:

[0072]

[0073] Due to scalar amplitude It is the dimensionless value obtained after normalization, and its statistical mean in this formula is... and standard deviation It is also a dimensionless parameter, therefore the calculated statistical fault triggering threshold Similarly, it is compatible with The dimensionless values ​​for direct comparison; k is a preset dimensionless sensitivity coefficient, whose source is set according to the statistical process control (SPC) criteria, for example, if Then this threshold corresponds to the Six Sigma criterion. Theoretically, this range covers 99.73% of health status fluctuations.

[0074] This threshold definition method, based on the statistical distribution of health cycles, replaces the traditional fixed empirical threshold; this threshold... It is derived from statistical learning of the equipment's historical data. It automatically incorporates the normal residual fluctuations that occur when the equipment is in a healthy state and responds to various complex operating conditions, which are reflected in... This enables the system to distinguish significant residuals caused by real physical degradation that exceed the normal statistical range with extremely high specificity, thereby effectively avoiding frequent false alarms caused by transient operating conditions and significantly improving the reliability of alarms.

[0075] Example 3

[0076] The global hydraulic condition vector includes upstream peak flow, downstream characteristic water level, gate opening sequence, and pump speed; the original mixed state dataset includes vibration signal vector, temperature of key components, and inlet and outlet pressure.

[0077] This embodiment specifically defines the physical components of the two sets of core vectors that the data acquisition unit needs to collect; this specific definition is the technical prerequisite for realizing the decoupling of working conditions, ensuring that the subsequent digital twin model has sufficient boundary conditions and verification data;

[0078] Global hydraulic condition vector This vector aims to comprehensively define the external boundary conditions driving hydraulic transients; in this embodiment, its components specifically include: upstream peak flow. Downstream characteristic water levels Gate opening sequence and pump unit speed These components together constitute all the driving inputs necessary for the runtime of computational fluid dynamics (CFD) or system hydraulics simulation models.

[0079] Original mixed-state dataset This dataset aims to comprehensively monitor critical internal responses of equipment caused by the aforementioned operating condition vectors; in this embodiment, its components specifically include: vibration signal vectors. Temperature of key components and import / export pressure For example, the pressure pulsation at the inlet and outlet of the pump; these data are the targets that the model needs to predict and verify.

[0080] By defining the input and output vectors so specifically, this embodiment ensures that the physical correspondence between cause and effect is clear and complete; it guarantees the subsequent stress baseline calculation unit. Having the complete boundary conditions required to perform physical simulations, it is possible to calculate high-fidelity hydraulic stress baselines. At the same time, it also ensured Sensor channels and The output channels are completely corresponding, enabling the vector difference calculation of the intrinsic residual analysis unit. It has clear physical meaning, laying a solid data foundation for subsequent fault decoupling and attribution.

[0081] Example 4

[0082] The process of updating the current health status vector of the equipment by the fault attribution unit is as follows: the output of the fault mode classifier is mapped to specific physical parameters in the digital twin reference model; at the same time, the scalar amplitude and duration are input into the preset damage accumulation model to quantitatively calculate the cumulative damage offset of the specific physical parameters.

[0083] The baseline vector corresponding to the preset absolute health state is combined with all identified cumulative damage offset vectors to dynamically maintain and output the current health state vector of the device.

[0084] This embodiment details how the fault attribution unit updates the current health state vector of the device after recognizing an intrinsic fault signal. The complete technical process; this is a core step that transforms abstract alarm signals into specific physical parameter offsets in a digital twin model;

[0085] In this embodiment, the update process couples two aspects: fault characterization and damage quantification.

[0086] Qualitative mapping: The physical calibration residual vector at the time the alarm is triggered. The input is fed into a fault mode classifier; the classifier's output will be mapped to a digital twin reference model. Specific physical parameters in For example, this fault type is mapped to a model. The parameter representing the blade geometry or the local hydraulic loss coefficient; this step determines which parameter should be modified in the digital twin model;

[0087] Quantitative calculation: The scalar amplitude at which the system will trigger an alarm. With duration These parameters are input into a pre-defined damage accumulation model, such as a fatigue crack propagation model based on Paris's law, or a wear model based on wear rate; this model is used to quantitatively calculate specific physical parameters. Cumulative damage offset This step determines how much the parameter should be modified.

[0088] Dynamic maintenance of state vectors: The system will preset the baseline state vector corresponding to the absolute health state. With all identified and calculated cumulative damage offset vectors Perform combinatorial operations; in this way, the system dynamically maintains and outputs the device's current health status vector. ;

[0089] This embodiment achieves dynamic quantification of device health status; it goes beyond traditional alarm levels, transforming fault signals into direct corrections to the physical parameters of the digital twin; this makes the updated health status vector... No longer an abstract health index, but a set of parameters for a digital twin model representing the current true physical degradation state of the device; this high-fidelity, degraded... Vectors are a core prerequisite for subsequent scheduling and avoidance units to perform accurate risk prediction.

[0090] Example 5

[0091] The weighted risk index is calculated as follows: Iterate through all key sensor channels; for each channel, obtain the model's predicted output value for that sensor, the preset safety alarm threshold, and the physical limit threshold; calculate the portion of the predicted output value that exceeds the safety alarm threshold, and divide it by the difference between the physical limit threshold and the safety alarm threshold to obtain the normalized degree of exceeding the limit.

[0092] Multiply the normalized exceedance level of each channel by the corresponding preset dimensionless risk weight, and sum the weighted results of all channels to obtain the dimensionless weighted risk index.

[0093] This embodiment specifies in detail the weighted risk index used by the scheduling avoidance unit to quantify operational risks. The specific calculation process; the purpose of this index is to unify the risk predictions of sensors with multiple dimensions and different units into a dimensionless total risk score that can be used for decision-making;

[0094] In this embodiment, the calculation process is performed during simulation prediction by the scheduling avoidance unit. Then start it up, as follows:

[0095] Channel traversal and threshold acquisition: As described in Example 6, the system traverses all key sensor channels. For example, vibration, temperature, pressure, etc.; for each channel The system retrieves three key values ​​from the configuration library:

[0096] The model's predicted output value for this sensor That is, the result of the operation obtained from the simulation;

[0097] Preset security alarm threshold Derived from equipment design specifications and industry safety standards, representing the reasonable upper limit for normal operation;

[0098] Physical limit threshold Also derived from design specifications, these represent the limits that could lead to permanent damage or failure.

[0099] Calculate the degree of exceeding the normalization limit: for each channel The system first calculates the predicted output value. Exceeding the security alarm threshold Part of (through) Calculate and use (Ensure the result is non-negative, i.e., 0 if it does not exceed the limit); divide the excess portion by the physical limit threshold. With security alarm threshold The difference between them (i.e.) This difference represents the safety margin range from alarm to damage;

[0100] This step (i.e.) The result is a normalized degree of over-limit; it is a dimensionless value representing how much of the safety margin the predicted risk encroaches upon.

[0101] Weighted summation: The system sums the values ​​of each channel. The above normalized excess limit multiplied by its corresponding preset dimensionless risk weight This weight This represents the different levels of importance of different sensor channels to the overall safety of the system; the value of this weight can be determined based on one or more of the following methods: based on expert knowledge and Failure Mode and Effects Analysis (FMEA), assign higher weights to sensor channels that are directly related to critical failure modes; based on historical data analysis, assign higher weights to sensor signals that have historically been more correlated with system failures.

[0102] Generate the overall index: Summing the weighted results of all channels yields a dimensionless weighted risk index. The complete calculation formula can be expressed as:

[0103]

[0104] To ensure robustness of the calculations, verification is required during system configuration for all channels. All satisfy the condition that the physical limit threshold is greater than the safety alarm threshold, i.e. To ensure the denominator is positive, and to prevent numerical instability due to a small difference between the two values, a very small positive constant can be added to the denominator in engineering implementation. To avoid the potential risk of division by zero;

[0105] This embodiment defines this weighted risk index, through... Normalization is performed so that risks of different physical dimensions can be measured and accumulated on a uniform scale; through weighting... The introduction of this technology allows risk assessment to focus more on key components that play a decisive role in overall safety; the generated dimensionless index It provides a clear, intuitive, and quantitatively accurate basis for scheduling decisions, ensuring the scientific nature and reliability of scheduling avoidance;

[0106] The preset risk threshold in this embodiment The definition method is as follows: This threshold aims to define the upper limit of acceptable risk. Its value can be calculated by performing risk simulations on a series of typical operational scenarios that have been assessed by experts as critically acceptable, and then calculating the corresponding weighted risk index. And take the statistical average or 95th percentile of these index values ​​as... This threshold represents the maximum level of risk the system is willing to take while ensuring safety.

[0107] Example 6

[0108] The scalar magnitude is a scalar value obtained by acquiring the L2 norm or Euclidean distance of the physical calibration residual vector.

[0109] This embodiment specifically defines the technical source of the scalar amplitude calculated by the intrinsic residual analysis unit; this scalar amplitude is the core basis for triggering fault determination;

[0110] In this embodiment, the physical calibration residual vector Rv(t) is a high-dimensional vector containing the residual components of all sensor channels. Since the physical dimensions of each component are different, in order to obtain a single dimensionless scalar value that can represent the overall magnitude of the residual, this embodiment specifies the scalar magnitude. Calculation method:

[0111] For residual vector Each component The statistical standard deviation collected and calculated during its healthy operating cycle is used. Normalization is performed to obtain the normalized residual components. To avoid instability in calculations due to a denominator of zero or too small, when the statistical standard deviation... When the value is less than a preset small positive threshold, it can be forcibly set to that threshold, or the normalized residual component of that channel can be... Set it directly to zero to enhance the robustness of the algorithm.

[0112] By obtaining the normalized physical calibration residual vector The scalar value is obtained by taking the L2 norm, and its calculation formula is as follows:

[0113]

[0114] in, It is the residual vector In the Components on each sensor channel This represents the total number of sensors. This normalization step effectively solves the problem of not being able to directly compare and sum the data from different sensor channels due to differences in dimensions and numerical ranges, ensuring the physical meaning and stability of the scalar amplitude.

[0115] in, It is the residual vector In the Components on each sensor channel;

[0116] Using the L2 norm as the calculation method for scalar amplitude, the L2 norm can comprehensively consider the residual contribution of all sensor channels, and it is very sensitive to significant deviations of any component in the vector; compared to the L1 norm or The L2 norm provides a smoother and more robust measure of overall deviation, which can most evenly reflect the degree of abnormal deviation in the multidimensional sensor space, ensuring the comprehensiveness and stability of subsequent fault triggering decisions based on the magnitude of this scalar.

[0117] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A system for analyzing the operational status of hydraulic engineering equipment, characterized in that, include: The data acquisition unit is used to acquire global hydraulic condition vectors and raw mixed state datasets in real time; The stress baseline calculation unit is used to perform physical simulation calculations on the global hydraulic condition vectors collected by the data acquisition unit based on the preset digital twin reference model and the preset absolute health state, so as to obtain the hydraulic stress baseline. The intrinsic residual analysis unit is used to perform vector difference calculation on the original mixed state dataset and the hydraulic stress baseline to obtain the physical calibration residual vector and scalar amplitude. It also performs discrimination processing on the scalar amplitude and the preset statistical fault trigger threshold to obtain the working condition stress signal or intrinsic fault signal. When an intrinsic fault signal is generated, the fault attribution unit is used to input the physical calibration residual vector at the time of triggering the alarm into a pre-trained fault mode classifier to identify the fault type, and combine the scalar amplitude and duration to calculate the cumulative damage offset through a preset damage accumulation model, thereby updating the current health status vector of the device. The scheduling and avoidance unit is used to respond to the received planned operation instructions, call the digital twin reference model, and input the global hydraulic condition vector collected in real time by the data acquisition unit, the planned operation instructions, and the current health status vector of the equipment updated or maintained by the fault attribution unit, to predict the operation consequences and calculate the weighted risk index. The weighted risk index is compared and analyzed with a preset risk threshold. When the weighted risk index exceeds the preset risk threshold, the planned operation instruction is rejected. The calculation process of the weighted risk index is as follows: traverse all key sensor channels; For each channel, obtain the model's predicted output value for the sensor, the preset safety alarm threshold, and the physical limit threshold; calculate the portion of the predicted output value that exceeds the safety alarm threshold, and divide it by the difference between the physical limit threshold and the safety alarm threshold to obtain the normalized degree of exceeding the limit. The normalized out-of-limit degree of each channel is multiplied by the corresponding preset dimensionless risk weight, and the weighted results of all channels are summed to obtain the dimensionless weighted risk index. The calculation formula is expressed as follows: ； in, For the first The preset dimensionless risk weights corresponding to each sensor channel represent the different levels of importance of different sensor channels to the overall safety of the system. To invoke the digital twin reference model and input the real-time acquired global hydraulic condition vector. Planned operation instructions and the device's current health status vector After that, the simulation results were obtained. The predicted output values ​​of each sensor; To preset the security alarm threshold; This is the physical limit threshold. It is an extremely small positive integer; to ensure the robustness of the calculation, it needs to be verified during system configuration for all channels. All satisfy the condition that the physical limit threshold is greater than the safety alarm threshold, i.e. To ensure that the denominator is positive; By obtaining the normalized physical calibration residual vector The scalar value is obtained by taking the L2 norm, and its calculation formula is as follows: ； in, This is the normalized physical calibration residual vector; It is the residual vector In the Components on each sensor channel; For the first The statistical standard deviation of each sensor channel during a healthy operating cycle; It represents the total number of sensors. Through a normalization process, the problem of incomparability and summation caused by differences in the dimensions and numerical ranges of data from different sensor channels is effectively solved, ensuring the physical meaning and stability of the scalar amplitude.

2. The system for analyzing the operational status of water conservancy engineering equipment according to claim 1, characterized in that, The statistical fault triggering threshold is defined as follows: Calculate the statistical distribution of the scalar amplitude within a known healthy operating cycle to obtain the mean and standard deviation; multiply the mean and standard deviation by a preset dimensionless sensitivity coefficient and add them together to obtain the statistical fault triggering threshold.

3. The system for analyzing the operational status of water conservancy engineering equipment according to claim 1, characterized in that, The global hydraulic condition vector includes upstream peak flow, downstream characteristic water level, gate opening sequence, and pump speed; the original mixed state dataset includes vibration signal vector, temperature of key components, and inlet and outlet pressure.

4. The system for analyzing the operational status of water conservancy engineering equipment according to claim 1, characterized in that, The process by which the fault attribution unit updates the current health status vector of the device is as follows: the output of the fault mode classifier is mapped to specific physical parameters in the digital twin reference model; at the same time, the scalar amplitude and duration are input into the preset damage accumulation model to quantitatively calculate the cumulative damage offset of the specific physical parameters.

5. The system for analyzing the operational status of water conservancy engineering equipment according to claim 4, characterized in that, The baseline vector corresponding to the preset absolute health state is combined with all identified cumulative damage offset vectors to dynamically maintain and output the current health state vector of the device.

Images