A one-button start adaptive control method for an aero-engine test stand

Abstract

This invention discloses a one-button start adaptive control method for aero-engine test benches, belonging to the field of aero-engine test technology. The method includes: responding to a one-button start command, collecting current environmental parameters and initial state parameters, and obtaining a database of historical successful start cases; performing multi-dimensional clustering based on the current operating conditions to determine clusters of similar successful cases; based on the start parameter sequences in the case clusters, using a digital twin model for multi-objective prediction and correction to generate an initial adaptive start control law; sending the control law to the field controller to execute the first ignition and speed increase, and collecting feedback operating data in real time; performing online rolling optimization of the control law based on the feedback data, and completing subsequent one-button continuous control after correction. This invention integrates historical cases and a digital twin model to achieve adaptive generation and online optimization of the control law, improving the start success rate and automation level.

Description

Technical Field

[0001] This invention relates to the field of aero-engine testing technology, specifically to a one-button start adaptive control method for an aero-engine test stand. Background Technology

[0002] The start-up process of an aero-engine test bench requires comprehensive consideration of various time-varying factors, such as ambient temperature, atmospheric pressure, lubricating oil temperature, and engine cylinder temperature. Any deviation in these parameters can lead to start-up failure or exceedances of critical parameters. Existing start-up control technologies typically employ pre-calibrated fixed control curves or simple lookup table strategies. Once the engine model and test bench hardware conditions are determined, the control parameters are rarely changed. This fixed control law is insufficient to cover the wide range of operating conditions that occur during actual test bench operation. Whenever there are seasonal changes, fluctuations in the test bench environment, or differences in the engine's initial thermal state, the same set of start-up parameters often causes problems such as excessively high exhaust temperatures, abnormal rotor acceleration rates, and excessive vibration amplitudes. Operators must repeatedly intervene manually, prolonging start-up preparation time and increasing test risks.

[0003] Existing solutions lack the ability to adaptively extract control strategies that match current operating conditions from historical successful experiences when dealing with multivariate coupled startup processes. They also lack the means to dynamically adjust based on real-time feedback during startup. This results in a large number of successful startup cases accumulated from historical test data not being effectively reused, with each startup still relying on inefficient fixed preset values. Furthermore, the control law cannot correct for instantaneous disturbances and model prediction deviations occurring during the startup phase online, making it difficult for the engine to maintain a smooth transition during acceleration. In severe cases, this can trigger over-temperature protection or vibration alarms, leading to startup interruption. Summary of the Invention

[0004] This paper presents a one-button start-up adaptive control method for aero-engine test benches. It generates an initial start-up control law that closely approximates the current operating conditions by multi-dimensional operating condition clustering of historical successful cases and prediction correction based on digital twin models. During execution, the control law is optimized online using real-time feedback data, enabling the engine to automatically complete a smooth start-up from ignition to idle under different environments and initial conditions, reducing manual intervention and lowering the risk of overheating and overvibration during the start-up process.

[0005] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides a one-button start adaptive control method for an aero-engine test stand, which can automatically match historical successful cases based on the current environment and the initial state of the engine, and combine digital twin model prediction correction and online rolling optimization to achieve one-button continuous control of the entire process from ignition to idle without manual intervention, significantly improving the start success rate and effectively suppressing the risks of over-temperature, over-vibration, and speed overshoot. The method includes the following steps: In response to the one-button start command, the system collects current environmental parameters and initial engine state parameters from the test bench, and obtains a database of historical successful start cases. Preferably, the system synchronously collects current ambient temperature, current atmospheric pressure, and current ambient humidity, as well as current engine oil temperature, current engine cylinder temperature, current fuel manifold pressure, and current initial rotor speed through a distributed sensor network on the test bench, serving as the current environmental parameters and initial engine state parameters. Using the engine model and test bench number as indexes, the system retrieves historical start records from the test bench's local historical database that match the current engine model and have a successful start result, thus obtaining the historical successful start case database. Each historical start record includes environmental parameters at the time of start, initial engine state parameters, a sequence of start control commands throughout the entire process, and a sequence of feedback states throughout the entire process. This step provides a comprehensive data foundation for subsequent operating condition matching and adaptive control.

[0006] Based on the current environmental parameters and engine initial state parameters, multi-dimensional operating condition clustering is performed on the historical successful start-up case database to determine multiple similar successful case clusters matching the current operating condition. Preferably, the current environmental parameters and engine initial state parameters are merged into a current operating condition feature vector; the environmental parameters at the start-up time and the engine initial state parameters are extracted for each historical start-up record and merged into a historical operating condition feature vector; the weighted Euclidean distance between the current operating condition feature vector and each historical operating condition feature vector is calculated, with the weight of each dimension determined by the sensitivity coefficient of each parameter to the start-up success rate; a predetermined number of historical start-up records with the highest ranking are selected as a candidate record set, and then divided according to three dimensions: the ambient temperature range, the lubricating oil temperature range, and the rotor initial speed range at the start-up time, to obtain multiple similar successful case clusters, each cluster corresponding to a set of range combinations. Through weighted sensitivity clustering, the start-up experience clusters closest to the current operating condition can be accurately selected, enhancing the operating condition adaptability of subsequent control laws.

[0007] Based on the startup parameter sequences from the multiple clusters of similar successful cases, a digital twin model is used to perform multi-objective prediction and correction of the startup process, generating an initial adaptive startup control law. Preferably, the full-process startup control command sequence and full-process feedback state sequence of all historical startup records are extracted from each cluster of similar successful cases and statistically averaged to obtain a cluster-averaged control command sequence and a cluster-averaged feedback state sequence. Using the current operating condition feature vector as input and multiple cluster-averaged control command sequences as the initial population, the startup process of each cluster-averaged control command sequence is simulated using a digital twin model to obtain a simulated feedback state sequence. The deviation sequence between the cluster-averaged feedback state sequence and the simulated feedback state sequence is calculated, and the corresponding cluster-averaged control command sequence is predicted and corrected based on the deviation sequence to obtain a corrected control command sequence for each cluster. The multiple corrected control command sequences are weighted and fused according to the sample size of each cluster to generate the initial adaptive startup control law. The digital twin model establishes thermodynamic, aerodynamic, and rotor dynamics simulation models based on the engine's physical characteristic equations. It utilizes full-process feedback state sequences from a historical database of successful startup cases to identify and calibrate key parameters in the simulation model, resulting in a digital twin model capable of simulating the startup process with high precision. The generated initial control law integrates multiple clusters of successful experiences and model prediction corrections, effectively compensating for differences in operating conditions and model errors, ensuring good adaptability from the initial setting of startup parameters.

[0008] The initial adaptive start-up control law is sent to the test bench controller to execute the engine's first ignition and speed increase operations, and the engine's feedback operating data is collected in real time during the execution process. Preferably, the initial adaptive start-up control law is divided into multiple sequentially arranged control command frames according to a preset time rhythm. Each control command frame includes a fuel flow setpoint, igniter on / off status, starter generator torque setpoint, and adjustable guide vane angle setpoint. The first control command frame is sent to the test bench controller, which controls the fuel regulating valve, ignition exciter, starter generator, and guide vane actuator to operate synchronously according to the first control command frame to execute the first ignition and speed increase operations. During the entire process of the engine accelerating from standstill to idle speed, subsequent control command frames are sent sequentially according to a preset time rhythm, and the engine's feedback operating data is recorded in real time by the test bench data acquisition system. The feedback operating data includes actual fuel flow, actual exhaust temperature, actual rotor speed, actual lubricating oil pressure, and actual vibration amplitude. This multi-loop synchronous sending and real-time acquisition mechanism ensures strict synchronization of the start-up sequence and comprehensive perception of the operating status.

[0009] Based on the feedback operation data, the initial adaptive start-up control law is optimized online to obtain a corrected adaptive start-up control law, and one-click continuous control is completed for the subsequent start-up phase based on this corrected control law. Preferably, the actual rotor speed in the feedback operation data is compared with the expected rotor speed at the corresponding moment in the initial adaptive start-up control law to obtain the speed deviation value; the actual exhaust temperature is compared with the expected exhaust temperature at the corresponding moment to obtain the temperature deviation value; and the actual vibration amplitude is compared with a preset vibration safety threshold to obtain the vibration margin value. Taking the current moment as the optimization start moment and the preset optimization time domain length as the optimization time domain end point, control command frames that have not yet been executed are extracted as the sequence of command frames to be optimized. The speed deviation value, temperature deviation value, and vibration margin value are used as state disturbance terms and are respectively superimposed on the sequence of command frames to be optimized. Based on the given values ​​of fuel flow rate and adjustable guide vane angle, a perturbed command frame sequence is obtained. Constrained by the engine rotor acceleration not exceeding a preset acceleration upper limit, exhaust temperature not exceeding a preset temperature upper limit, and vibration amplitude not exceeding a vibration safety threshold, and with the optimization objective of minimizing the difference between the desired rotor speed and the predicted rotor speed at the end of the optimization time domain, the given values ​​of fuel flow rate and adjustable guide vane angle in the perturbed command frame sequence are iteratively optimized to obtain a corrected future time period control command frame sequence. This corrected sequence replaces the original unexecuted frames, forming the corrected adaptive start-up control law. The first unexecuted control command frame in the corrected control law is issued and executed. After collecting the latest feedback operating data, subsequent unexecuted frames are updated again. This process of issuing and updating is repeated until the engine rotor speed reaches the idle speed setpoint and the exhaust temperature stabilizes within the allowable idle temperature range. Preferably, when the rotor speed reaches a preset proportion of the idle speed setpoint, the fuel flow command is switched to an idle fuel flow command corrected based on the current ambient atmospheric pressure. After the switch, rotor speed and exhaust temperature are continuously monitored. When the rotor speed remains within a preset fluctuation range of the idle speed setpoint and the exhaust temperature remains within the allowable idle temperature range for a duration exceeding a preset stabilization time threshold, the one-button continuous control for the subsequent start-up phase is considered complete. By continuously optimizing and correcting control commands online, this method can actively suppress speed fluctuations, excessive exhaust temperature, and abnormal vibrations during the start-up process, ensuring a smooth and safe start-up to the idle stable state.

[0010] The technical effects and advantages provided by the present invention in the above technical solution are as follows: Multidimensional operating condition clustering extracts multiple clusters of similar successful cases from historical start-up cases, which best match the current ambient temperature, lubricating oil temperature, and rotor initial speed. The similarity is measured using a weighted Euclidean distance between the historical and current operating condition feature vectors, with the weights of different parameters determined by their sensitivity coefficients to start-up success rates. This ensures that the selected case clusters closely approximate current physical conditions in terms of start-up response characteristics. When generating the initial adaptive start-up control law, a cluster-average control command sequence and a cluster-average feedback state sequence are extracted for each similar case cluster. A digital twin model is used to simulate the start-up process of the cluster-average control command sequence, obtaining the simulated feedback state sequence. The deviation sequence between the cluster-average feedback state sequence and the simulated feedback state sequence is then used to predict and correct the corresponding cluster-average control command sequence. This eliminates model mismatch errors caused by pure historical statistical averaging, ensuring that the initial control law incorporates high-precision predictions of engine thermodynamics and aerodynamic characteristics before actual ignition. This effectively suppresses initial speed overshoot and sharp rise in exhaust temperature caused by differences in environmental parameters and engine initial conditions.

[0011] The online rolling optimization process uses the deviations between the actual rotor speed and the desired speed, the deviations between the actual exhaust temperature and the desired temperature, and the margin of the actual vibration amplitude relative to the vibration safety threshold as state disturbance terms. These are superimposed on the fuel flow setpoint and the adjustable guide vane angle setpoint in the control command frames that have not yet been executed, forming a disturbed command frame sequence. Iterative optimization is performed with constraints of the upper limit of rotor acceleration, the upper limit of exhaust temperature, and the vibration safety threshold, aiming to minimize the difference between the desired speed and the predicted speed at the end of the optimization time domain. This directly corrects subsequent control command frames within continuous time cycles. This time-domain rolling optimization method based on multi-source real-time feedback deviation can dynamically compensate for transient deviations caused by factors such as fuel quality fluctuations, sensor drift, and bench airflow disturbances during startup. It can continuously adjust the fuel flow and guide vane angle without interrupting the startup process, ensuring a smooth transition of the engine rotor speed to idle speed, and keeping the exhaust temperature and vibration amplitude within safe boundaries, avoiding startup failure due to reaching preset protection thresholds. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0013] Figure 1 This is a flowchart of the one-button start adaptive control method for aero-engine test benches; Figure 2This is a flowchart of the one-button start data acquisition and historical data retrieval process for the test bench; Figure 3 This is a flowchart illustrating the method for dividing clusters of similar success cases. Figure 4 This is the flowchart for generating the initial adaptive startup control law. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0015] See Figure 1 This invention provides a one-button start adaptive control method for an aero-engine test stand. The method includes: responding to a one-button start command by collecting current environmental parameters of the test stand and initial engine state parameters, and obtaining a historical database of successful start cases; performing multi-dimensional operating condition clustering on the historical database of successful start cases based on the current environmental parameters and initial engine state parameters to determine multiple clusters of similar successful cases matching the current operating conditions; based on the start parameter sequences in the multiple clusters of similar successful cases, using a digital twin model to perform multi-objective prediction and correction of the start-up process, generating an initial adaptive start control law; sending the initial adaptive start control law to the test stand field controller to execute the engine's first ignition and speed increase operations, and collecting real-time feedback operating data of the engine during execution; performing online rolling optimization of the initial adaptive start control law based on the feedback operating data to obtain a corrected adaptive start control law, and completing one-button continuous control of subsequent start-up phases based on the corrected adaptive start control law.

[0016] Example 1 In specific implementation, please refer to Figure 2Upon receiving the one-button start command from the operator via the human-machine interface, the central control unit of the test bench immediately triggers the data acquisition process. The test bench's distributed sensor network consists of an environmental parameter sensor group located in the intake area and an engine status sensor group installed at various measuring points on the engine body. Both the environmental parameter sensor group and the engine status sensor group are connected to the same time synchronization server, achieving microsecond-level clock synchronization via the IEEE 1588 precision time protocol. At the same moment the synchronization acquisition trigger signal is received, the platinum resistance temperature sensor in the environmental parameter sensor group acquires the current ambient temperature, the silicon resonant atmospheric pressure sensor acquires the current ambient atmospheric pressure, and the capacitive humidity sensor acquires the current ambient humidity; the lubricating oil temperature thermocouple in the engine status sensor group acquires the current lubricating oil temperature, the cylinder temperature thermocouple acquires the current engine cylinder temperature, the fuel manifold pressure transmitter acquires the current fuel manifold pressure, and the magnetoelectric speed sensor acquires the current rotor initial speed. These seven acquired parameters are uniformly timestamped and packaged into data frames for current environmental parameters and engine initial status parameters, which are then transmitted to the central control unit via the test bench's real-time data bus.

[0017] While collecting current environmental parameters and initial engine state parameters, the central control unit initiates a search request to the local historical database of the test bench, using the engine model and test bench number as a composite index. The local historical database is a relational database containing a historical start-up record table with the engine model and test bench number as the composite primary key. Each record in the historical start-up record table corresponds to a complete start-up process and includes a start-up result identifier field. The search criteria are: the engine model field equals the current engine model, the test bench number field equals the current test bench number, and the start-up result identifier field is set to "success". All historical start-up records that meet the filter criteria are extracted, forming a historical start-up success case database. Each historical start-up record contains the following fields: ambient temperature at start-up, ambient atmospheric pressure at start-up, ambient humidity at start-up, lubricating oil temperature at start-up, engine cylinder temperature at start-up, fuel manifold pressure at start-up, initial rotor speed at start-up, and a full-process start-up control command sequence and a full-process feedback state sequence stored at a uniform time interval. Each time step in the full-process start-up control command sequence records the fuel flow setpoint, igniter on / off status, starter generator torque setpoint, and adjustable guide vane angle setpoint; each time step in the full-process feedback status sequence records the actual fuel flow, actual exhaust temperature, actual rotor speed, actual lubricating oil pressure, and actual vibration amplitude.

[0018] Example 2 In specific implementation, please refer to Figure 3After acquiring current environmental parameters, engine initial state parameters, and a database of historical successful start-ups, the central control unit constructs a working condition feature vector. The central control unit arranges the current ambient temperature, current atmospheric pressure, current ambient humidity, current lubricating oil temperature, current engine cylinder temperature, current fuel manifold pressure, and current rotor initial speed in a preset order to form a seven-dimensional current working condition feature vector, denoted as... For each historical start-up record in the historical successful start-up case database, the central control unit extracts the ambient temperature, atmospheric pressure, humidity, lubricating oil temperature, engine cylinder temperature, fuel manifold pressure, and initial rotor speed at the start-up time stored in that record, and then processes these data according to... Arranged in the same preset order, these form the historical operating condition feature vector for that historical startup record. The historical operating condition feature vectors corresponding to all historical startup records in the historical successful startup case database constitute the historical operating condition feature vector set.

[0019] In some embodiments, the central control unit calculates the relationship between each historical operating condition feature vector in the historical operating condition feature vector set and the current operating condition feature vector. The weighted Euclidean distance between them. The formula for calculating the weighted Euclidean distance is: in, Represents the feature vector of the current operating condition. With the Historical startup record historical operating condition feature vector The weighted Euclidean distance between them; For the dimension index number, The value range is an integer from 1 to 7, corresponding to seven dimensions: ambient temperature, ambient atmospheric pressure, ambient humidity, lubricating oil temperature, engine cylinder temperature, fuel main pressure, and rotor initial speed. Represents the feature vector of the current operating condition. In the Values ​​in each dimension; Indicates the first Historical startup record historical operating condition feature vector In the Values ​​in each dimension; For the first The weight corresponding to the dimension, which is determined by the . The sensitivity coefficients of the parameters corresponding to each dimension to the startup success rate are determined. The higher the sensitivity coefficient, the higher the corresponding weight. The larger the value, the higher the sensitivity coefficient. The sensitivity coefficient is pre-calculated through historical data analysis: the central control unit extracts sample data from historical successful startup case databases and historical failed startup case databases, and fits a logistic regression curve between each parameter dimension and the success or failure of the startup. The absolute value of the regression coefficient corresponding to that parameter in the logistic regression curve is used as the sensitivity coefficient of that parameter to the startup success rate. The sensitivity coefficients corresponding to the above seven parameters are normalized and then directly assigned to their corresponding weights. This makes all seven weights The sum is 1.

[0020] In practice, the central control unit obtains the weighted Euclidean distance between the current operating condition feature vector and the historical operating condition feature vector of each historical startup record. Then, it sorts all historical startup records in ascending order of their weighted Euclidean distance and selects a predetermined number of these records as a candidate record set. The predetermined number is set based on the total capacity of the historical startup success case database. When the total number of historical startup records in the database exceeds 500, the predetermined number is 50; when the total number of historical startup records in the database does not exceed 500, the predetermined number is an integer value rounded up to 10% of the total number of historical startup records.

[0021] Optionally, the central control unit divides each historical startup record in the candidate record set according to three dimensions: the ambient temperature range at startup, the lubricating oil temperature range at startup, and the rotor initial speed range at startup. The ambient temperature range is divided into three zones: low temperature, normal temperature, and high temperature. The low temperature zone corresponds to an ambient temperature below -10 degrees Celsius, the normal temperature zone corresponds to an ambient temperature between -10 degrees Celsius and 40 degrees Celsius, and the high temperature zone corresponds to an ambient temperature above 40 degrees Celsius. The lubricating oil temperature range is divided into two zones: cold start zone and warm-up zone. The cold start zone corresponds to a lubricating oil temperature below 30 degrees Celsius, and the warm-up zone corresponds to a lubricating oil temperature not lower than 30 degrees Celsius. The rotor initial speed range is divided into two zones: stationary zone and pre-rotation zone. The stationary zone corresponds to an initial rotor speed of 0 rpm, and the pre-rotation zone corresponds to an initial rotor speed greater than 0 rpm. The central control unit iterates through each historical startup record in the candidate record set. Based on the intervals that the ambient temperature, lubricating oil temperature, and initial rotor speed at startup fall into at that startup time, the historical startup record is divided into corresponding groups. Each group is a cluster of similar successful cases, and each cluster corresponds to a three-dimensional interval combination consisting of an ambient temperature interval, a lubricating oil temperature interval, and an initial rotor speed interval. After the division, the historical startup records in the candidate record set are divided into several clusters of similar successful cases. Each cluster contains several historical startup records that fall within the same interval combination in the three dimensions of ambient temperature, lubricating oil temperature, and initial rotor speed. These clusters of similar successful cases are the multiple clusters of similar successful cases that match the current operating condition.

[0022] Example 3 In specific implementation, please refer to Figure 4The process of generating the initial adaptive start-up control law begins by extracting historical start-up record data from multiple identified clusters of similar successful cases. For each cluster of similar successful cases, the central control unit traverses all historical start-up records contained within that cluster, extracting the control command values ​​at each uniform time point during the start-up process from the complete start-up control command sequence of each historical start-up record. The control command values ​​include four components: fuel flow setpoint, ignition on / off state, starter generator torque setpoint, and adjustable guide vane angle setpoint. For each time tick point in the startup process, the central control unit performs an arithmetic average of the fuel flow setpoint components of all historical startup records within the similar successful case cluster at that time tick point, takes the majority vote of the ignition on / off status components of all historical startup records within the similar successful case cluster at that time tick point, performs an arithmetic average of the starter generator torque setpoint components of all historical startup records within the similar successful case cluster at that time tick point, and performs an arithmetic average of the adjustable guide vane angle setpoint components of all historical startup records within the similar successful case cluster at that time tick point. The statistical processing results of the above four components constitute a control command average vector for that similar successful case cluster at that time tick point. Arranging the control command average vectors at all time tick points in the startup process in chronological order yields the cluster average control command sequence corresponding to that similar successful case cluster. The central control unit extracts the feedback state value at each time step from the entire feedback state sequence of all historical startup records within the cluster of similar successful cases, using the same time step alignment method. The feedback state value includes five components: actual fuel flow, actual exhaust temperature, actual rotor speed, actual lubricating oil pressure, and actual vibration amplitude. An arithmetic mean is calculated for each component at each time step to obtain an average feedback state vector. These average feedback state vectors are then arranged chronologically to obtain the cluster-averaged feedback state sequence corresponding to the cluster of similar successful cases. This process is repeated for each cluster of similar successful cases, resulting in multiple cluster-averaged control command sequences and multiple cluster-averaged feedback state sequences, each corresponding to one of the clusters.

[0023] In practical implementation, the digital twin model is a high-precision simulation model of the engine startup process driven by both physical mechanisms and data. This digital twin model is built offline and deployed in the central control unit before generating the initial adaptive startup control law. The construction method of the digital twin model is as follows: Based on the component-level physical characteristic equations of the engine, a thermodynamic, aerodynamic, and rotor dynamics simulation model is established, including a compressor sub-model, a combustion chamber sub-model, a turbine sub-model, and a rotor dynamics sub-model. The compressor sub-model employs a stage-by-stage characteristic line interpolation method to calculate the outlet total temperature, total pressure, and flow rate based on the inlet total temperature, total pressure, rotor speed, and adjustable guide vane angle. The combustion chamber sub-model, based on the energy conservation equation, calculates the outlet total gas temperature and total pressure based on the inlet air flow rate, temperature, and fuel flow rate. The turbine sub-model uses a similar stage-by-stage characteristic line interpolation method to calculate the outlet total temperature, total pressure, and output torque based on the inlet total gas temperature, total pressure, and rotor speed. The rotor dynamics sub-model establishes a dynamic balance equation between rotor acceleration and turbine output torque, compressor torque consumption, generator applied torque, and friction torque based on Newton's second law. These sub-models are coupled through working fluid and torque transmission relationships to form a complete startup process simulation model. This simulation model includes several key parameters to be identified, including compressor stage efficiency correction coefficients, combustion chamber heat loss coefficients, turbine stage efficiency correction coefficients, and rotor moment of inertia.

[0024] In practice, the central control unit extracts the full-process feedback state sequence and corresponding full-process start-up control command sequence of all historical start-up records from the historical successful start-up case database to construct a training dataset. The central control unit uses the full-process start-up control command sequence of each historical start-up record as input to the simulation model, and simultaneously uses the environmental parameters at the start-up time and the engine initial state parameters of that historical start-up record as initial conditions input to the simulation model, driving the simulation model to run and outputting the simulation full-process state sequence. The central control unit uses the weighted sum of the root mean square errors between the actual rotor speed, actual exhaust temperature, and actual fuel flow rate in the simulation full-process state sequence and the corresponding full-process feedback state sequence in the historical start-up records as the loss function. It then employs an adaptive moment estimation optimization algorithm to repeatedly adjust the compressor stage efficiency correction coefficient, combustion chamber heat loss coefficient, turbine stage efficiency correction coefficient, and rotor moment of inertia value until the loss function converges to below a preset threshold. After convergence, the key parameter values ​​are substituted into the simulation model to obtain the identified and calibrated digital twin model. This digital twin model can predict the time history of rotor speed, exhaust temperature, fuel flow, lubricating oil pressure and vibration amplitude during the engine start-up process based on any given control command sequence and initial operating conditions.

[0025] In practical implementation, the current operating condition feature vector is used as the initial operating condition input for the digital twin model. The cluster-averaged control command sequence of a similar successful case cluster is used as the control command sequence to be simulated and input into the digital twin model. The digital twin model performs simulation calculations step by step according to the startup process time cycle, and outputs the simulation feedback state sequence under the action of the cluster-averaged control command sequence of the similar successful case cluster. The simulation feedback state sequence includes the simulated fuel flow rate, simulated exhaust temperature, simulated rotor speed, simulated lubricating oil pressure, and simulated vibration amplitude at each time cycle point. The above simulation operation is performed for the cluster-averaged control command sequence of each similar successful case cluster, and the central control unit obtains multiple simulation feedback state sequences that correspond one-to-one with multiple similar successful case clusters.

[0026] In practical implementation, for each cluster of similar successful cases, the central control unit subtracts each component of the cluster-averaged feedback state sequence of the cluster from the corresponding component in the simulation feedback state sequence at the same time tick point to obtain the deviation sequence. The deviation sequence also contains five components: fuel flow deviation component, exhaust temperature deviation component, rotor speed deviation component, lubricating oil pressure deviation component, and vibration amplitude deviation component. The central control unit uses the deviation sequence as a correction signal to predictively correct the cluster-averaged control command sequence. The specific correction method is as follows: at each time tick point, the rotor speed deviation component in the deviation sequence is multiplied by a first correction gain coefficient and then superimposed on the fuel flow setpoint component at the corresponding time in the cluster-averaged control command sequence; the exhaust temperature deviation component in the deviation sequence is multiplied by a second correction gain coefficient and then superimposed on the adjustable guide vane angle setpoint component at the corresponding time in the cluster-averaged control command sequence. The ignition on / off state and the starter generator torque setpoint remain unchanged during the predictive correction. Both the first and second correction gain coefficients are constants obtained through offline tuning based on engine open-loop step response test data. The first correction gain coefficient is set to 0.02 kg / s per revolution per minute, and the second correction gain coefficient is set to 0.5 degrees per degree Celsius. After correction operations at all time points, the corrected control command sequence corresponding to the cluster of similar successful cases is obtained. The above predictive correction operation is performed on each cluster of similar successful cases, and the central control unit obtains multiple corrected control command sequences corresponding one-to-one with the multiple clusters of similar successful cases.

[0027] In practice, the central control unit weighted and fused multiple corrected control command sequences corresponding to several similar successful case clusters to generate an initial adaptive start-up control law. The calculation formula for the weighted fusion is as follows: in, This indicates the initial adaptive startup control law during the startup process. Time beat point The four-dimensional control command vector at any given time includes the fuel flow setpoint, igniter on / off state, starter / generator torque setpoint, and adjustable guide vane angle setpoint. This represents the total number of clusters of similar successful cases that match the current working condition; This is the index number of the cluster of similar success cases. Take from 1 ; Indicates the first The number of historical startup records contained in a cluster of similar success cases is the sample size of that cluster of similar success cases. Indicates the first The corrected control command sequence corresponding to the cluster of similar successful cases is in the... Time beat point A four-dimensional control command vector at each moment. The central control unit controls each time step during the startup process. Perform the above weighted fusion calculation, and calculate all time points. Arranged in chronological order, the complete initial adaptive startup control law is obtained.

[0028] Example 4 In practice, after the central control unit generates the initial adaptive start-up control law, it converts the law into a data format suitable for execution by the test bench's field controller and distributes it via the industrial Ethernet bus. The preset time interval is a fixed control cycle uniformly adopted by the engine's full authority digital electronic control system, with a value of 20 milliseconds. Using this fixed control cycle as an interval, the central control unit divides the time-continuous sequence of control command vectors in the initial adaptive start-up control law into multiple sequentially arranged control command frames. Each control command frame corresponds to a control cycle time, with frame numbers incrementing sequentially from zero. The frame number multiplied by the fixed control cycle gives the absolute time of that control command frame relative to the start-up zero time. Each control command frame contains four control command components: fuel flow rate setpoint, ignition on / off status, starter / generator torque setpoint, and adjustable guide vane angle setpoint. These four control command components are encoded in big-endian order with a fixed byte length in the control command frame. The fuel flow rate setpoint is encoded using a 32-bit signed floating-point number in kilograms per hour. The ignition on / off status is encoded using a single-byte Boolean value. The starter / generator torque setpoint is encoded using a 32-bit signed floating-point number in Newton-meters. The adjustable guide vane angle setpoint is encoded using a 32-bit signed floating-point number in degrees.

[0029] In practical implementation, the central control unit sends the first control command frame with frame number zero to the test bench field controller via the industrial Ethernet bus. The test bench field controller is an embedded programmable automatic controller based on a real-time operating system, which internally runs closed-loop regulation tasks and command parsing tasks. After receiving the first control command frame, the command parsing task of the test bench field controller performs cyclic redundancy check on the control command frame. After the check passes, the control command frame is unpacked into fuel flow setpoint, ignition on / off status, starter generator torque setpoint, and adjustable guide vane angle setpoint according to the encoding rules. The test bench field controller writes the fuel flow setpoint into the target value register of the proportional-integral-derivative controller in the inner loop of the fuel regulating valve position. The proportional-integral-derivative controller in the inner loop of the fuel regulating valve position generates a pulse width modulation drive signal to drive the torque motor of the fuel regulating valve based on the deviation between the target value and the actual valve position fed back by the linear variable differential transformer installed on the valve stem of the fuel regulating valve. This drives the valve core of the fuel regulating valve to move to the opening position corresponding to the fuel flow setpoint, thereby regulating the fuel supply to the engine combustion chamber. The test bench field controller directly writes the ignition on / off status into the digital output channel. When the ignition is on, the digital output channel outputs a high level to drive the solid-state relay of the ignition actuator to close. The high-voltage capacitor inside the ignition actuator is charged and then releases a high-voltage pulse to the ignition nozzle, achieving ignition in the engine combustion chamber. The test bench field controller writes the starter generator torque setpoint into the torque command register of the starter generator vector control driver. The starter generator vector control driver controls the orthogonal axis component of the stator current vector of the starter generator according to the torque command, so that the starter generator outputs torque corresponding to the starter generator torque setpoint, which drives the engine high-pressure rotor to rotate through the accessory gearbox. The test bench's on-site controller writes the setpoint value for the adjustable guide vane angle into the target value register of the guide vane actuator position loop proportional-integral-derivative controller. Based on the deviation between the target value and the actual displacement fed back by the magnetostrictive displacement sensor mounted on the actuator piston rod, the guide vane actuator position loop proportional-integral-derivative controller drives the electro-hydraulic servo valve to control the hydraulic circuit, pushing the guide vane actuator piston to move. This synchronously changes the angle of each stage of the adjustable guide vane through a linkage mechanism. The four types of actuators—fuel regulating valve, ignition exciter, starter generator, and guide vane actuator—operate synchronously under the coordination of the test bench's on-site controller, jointly performing the engine's initial ignition and speed increase operations.

[0030] In practice, during the entire process of the engine accelerating from standstill to idle speed, the central control unit sequentially sends subsequent control command frames with increasing frame numbers to the test bench field controller according to a fixed control cycle. The test bench field controller parses and executes each received control command frame in real time. Parallel to the command issuance process, the test bench data acquisition system uses a fixed control cycle as its sampling period, synchronously recording the engine's feedback operating data in real time through analog and digital acquisition channels. The actual fuel flow rate is measured in real time by a Coriolis mass flow meter installed at the fuel manifold inlet. The Coriolis mass flow meter outputs a current signal of 4 mA to 20 mA, which, after signal conditioning, is converted into an engineering value of the actual fuel flow rate in kilograms per hour by an analog-to-digital converter. The actual exhaust temperature is measured in real time by six K-type armored thermocouples evenly distributed circumferentially along the exhaust cross-section at the turbine outlet. The voltage signals of the six K-type armored thermocouples are arithmetically averaged after cold junction compensation and multiplexing, and converted into an engineering value of the actual exhaust temperature in degrees Celsius. The actual rotor speed is measured in real time by a high-precision magnetoelectric speed sensor installed at the high-pressure rotor shaft end. The pulse frequency signal generated by the rotor speed measuring gear plate sensed by the high-precision magnetoelectric speed sensor is converted into an analog voltage by a frequency-to-voltage conversion circuit and then acquired by an analog-to-digital converter. Finally, it is converted into the actual rotor speed engineering value, in revolutions per minute (rpm). The actual lubricating oil pressure is measured in real time by a sputtered diaphragm pressure transmitter installed on the lubricating oil supply manifold. The sputtered diaphragm pressure transmitter outputs a current signal of 4 mA to 20 mA, which is converted into the actual lubricating oil pressure engineering value by an analog-to-digital converter after signal conditioning, in kilopascals (kPa). The actual vibration amplitude is measured in real time by piezoelectric accelerometer vibration sensors installed at three measuring points on the engine front casing, intermediate casing, and rear casing. The charge signals of the three piezoelectric accelerometer vibration sensors are converted into voltage signals by a charge amplifier and then acquired by an analog-to-digital converter. The maximum value of the three channels is taken as the actual vibration amplitude engineering value, in millimeters per second (mm / s). The test bench data acquisition system packages the five engineering values ​​collected in each control cycle and the corresponding control cycle number into a feedback data frame, which is then transmitted back to the central control unit via the industrial Ethernet bus, forming the feedback operation data for the entire engine start-up process.

[0031] In practice, after receiving the latest batch of feedback operating data in each control cycle, the central control unit immediately initiates an online rolling optimization process to revise the initial adaptive start-up control law. The central control unit extracts the actual rotor speed corresponding to the current control cycle from the feedback operating data, and simultaneously indexes the expected rotor speed corresponding to the current control cycle from the initial adaptive start-up control law. This expected rotor speed is the predicted value of the future state calculated using a digital twin model during the weighted fusion calculation phase of the initial adaptive start-up control law generation stage, and is recorded in the state trajectory of the initial adaptive start-up control law. The central control unit calculates the difference between the actual rotor speed and the expected rotor speed to obtain the speed deviation value.

[0032] In practical implementation, the central control unit extracts the actual exhaust temperature corresponding to the current control cycle from the feedback operation data and indexes the desired exhaust temperature corresponding to the current control cycle from the initial adaptive start-up control law. The desired exhaust temperature also originates from the state trajectory record of the initial adaptive start-up control law. The difference between the actual exhaust temperature and the desired exhaust temperature is calculated to obtain the temperature deviation value. The central control unit extracts the actual vibration amplitude corresponding to the current control cycle from the feedback operation data and reads the preset vibration safety threshold value from the central control unit's configuration parameter memory. This vibration safety threshold value is the upper limit of the allowable vibration amplitude during the start-up phase specified by the engine manufacturer in the engine maintenance manual, and is set to 75 mm / s. The difference between the vibration safety threshold value and the actual vibration amplitude is calculated to obtain the vibration margin value. When the actual vibration amplitude has reached or exceeded the vibration safety threshold value, the vibration margin value is zero or negative. In this case, the rolling time-domain optimization algorithm will prioritize satisfying the vibration constraints.

[0033] In practical implementation, after calculating the three deviation values—speed deviation, temperature deviation, and vibration margin—the central control unit uses a rolling time-domain optimization algorithm to correct the control command frames that have not yet been executed in the initial adaptive start-up control law. The central control unit takes the end time of the current control cycle as the optimization start time, denoted as _____. The central control unit reads the preset optimization time domain length from the configuration parameter memory. The optimization time domain length represents the future time span covered by one rolling optimization, and its value is 40 fixed control cycles, or 800 milliseconds. The optimization time domain end point is the optimization start time. With the optimized time domain length determined, let the optimized time domain endpoint be denoted as . The central control unit extracts the timestamp from the initial adaptive startup control law at the optimization start time. With optimization of the time domain endpoint All control command frames between them form a sorted set of command frames, which serves as the sequence of command frames to be optimized.

[0034] In practice, the central control unit records the speed deviation value as... The unit is revolutions per minute (rpm), and the temperature deviation value is recorded as... The unit is Celsius, and the vibration margin value is denoted as... The unit is millimeters per second. The central control unit will display the speed deviation value. The fuel flow disturbance component is obtained by multiplying by the fuel flow disturbance mapping coefficient. The value range of the fuel flow disturbance mapping coefficient is obtained through engine step response testing and is set to 0.15 kg / h / rpm. The central control unit will then calculate the temperature deviation value. and vibration margin value The linear combination of these values ​​is used as the input for the guide vane disturbance mapping, specifically in the form of: the temperature deviation value... Multiply by the temperature disturbance factor of 0.2 degrees per degree Celsius, and add the vibration margin value. Multiplying by a vibration disturbance coefficient of 0.1 degrees per millimeter per second yields the guide vane angle disturbance component, in degrees. The central control unit iterates through each control command frame in the sequence of command frames to be optimized, adding the fuel flow disturbance component to the fuel flow setpoint component and the guide vane angle disturbance component to the adjustable guide vane angle setpoint component. The ignition on / off state and starter torque setpoint components in the control command frames remain unchanged, thus obtaining the perturbed command frame sequence. This perturbed command frame sequence serves as the initial solution for the rolling time-domain optimization algorithm.

[0035] In practical implementation, the central control unit starts with the perturbed instruction frame sequence and uses a sequential quadratic programming algorithm for iterative optimization. Sequential quadratic programming is a numerical algorithm for solving constrained nonlinear optimization problems. In each iteration, it linearizes the nonlinear optimization problem into a quadratic programming subproblem at the current iteration point. By solving this subproblem, the search direction and step size are determined, and the iteration point is updated until the convergence condition is met. The complete mathematical description of rolling time-domain optimization is as follows.

[0036] Optimization variables: The optimization variables for the rolling time-domain optimization problem are those starting from the optimization start time. To the end of the optimized time domain The sequence of fuel flow rate setpoints and the sequence of adjustable guide vane angle setpoints at each control cycle time will be used to determine the first... The fuel flow rate setpoint at each control cycle time is denoted as , will the The adjustable guide vane angle setpoint at each control cycle is denoted as . , From 0 ,in To optimize the number of control cycles contained in the time domain, It equals the optimized time domain length divided by the fixed control period, i.e. .

[0037] Constraints: Three hard constraints are applied during the optimization process. The first constraint is the rotor acceleration constraint: in the... At any given control cycle, the engine rotor acceleration predicted by the digital twin model, after low-pass filtering, must not exceed the preset acceleration upper limit, which is set at 50 rpm per control cycle. The second constraint is the exhaust temperature constraint: at the... At any given control cycle time, the exhaust temperature predicted by the digital twin model must not exceed the preset temperature upper limit. The preset temperature upper limit is the exhaust temperature redline value corresponding to the engine type during the start-up phase, which is set to 850 degrees Celsius. The third constraint is a vibration constraint: at the... At any given control cycle, the vibration amplitude predicted by the digital twin model must not exceed the preset value of the vibration safety threshold of 75 mm / s.

[0038] Optimization objective: The objective function of optimization is to minimize the time-domain endpoint. The absolute value of the difference between the desired rotor speed and the predicted rotor speed at the desired rotor speed, where the desired rotor speed is determined by the state trajectory in the initial adaptive start-up control law at the end of the optimization time domain. The expected value at the given point is determined, and the predicted rotor speed is simulated by a digital twin model using the sequence of optimization variables and the current engine state as inputs, up to the optimization time domain endpoint. get.

[0039] In its implementation, the sequential quadratic programming algorithm uses a digital twin model to simulate the current sequence of optimization variables in each iteration. The digital twin model uses the latest feedback data from the current control cycle as its initial state, including the actual rotor speed, actual exhaust temperature, actual fuel flow rate, actual lubricating oil pressure, and actual vibration amplitude at the current moment. It then sequentially calculates the predicted rotor speed, predicted exhaust temperature, and predicted vibration amplitude for each control cycle within the optimization time domain. Based on the predicted output of the digital twin model, the sequential quadratic programming algorithm calculates the objective function value and constraint violation amount, constructs the Hessian matrix approximation of the Lagrangian function at the current iteration point, and transforms the nonlinear optimization problem into a quadratic programming subproblem with linear constraints. The active set method is then used to solve this quadratic programming subproblem, yielding the desired result. and The search direction is determined, and a line search is performed along the search direction to find the step size that makes the evaluation function decrease and satisfies the constraints. All steps are then updated. and The value of the objective function is then repeated. When the change in the objective function value between two consecutive iterations is less than the preset convergence threshold of 0.01, the sequential quadratic programming algorithm converges, and the final value is output. and The sequence. The central control unit will converge all the obtained sequences. The value is written to the fuel flow setpoint component of the corresponding control command frame, and all The value is written into the adjustable guide vane angle setpoint component of the corresponding control command frame. The igniter on / off state and the starter generator torque setpoint use the values ​​from the starting point of the iterative optimization, thus assembling a corrected future time period control command frame sequence.

[0040] In practice, the central control unit compares the revised future time period control command frame sequence with the initial adaptive start-up control law to locate the timestamp in the initial adaptive start-up control law at the optimization start time. For any subsequent raw control command frames that have not yet been executed, these raw control command frames are deleted from the control command buffer of the central control unit. Then, control command frames from the corrected future time period control command frame sequence are inserted into the control command buffer one-to-one according to their timestamps, completing the replacement operation. After the replacement operation, all control command frames in the control command buffer from the current control cycle onwards are control command frames corrected by the latest round of online rolling optimization. The central control unit uses the entire sequence of control command frames in the control command buffer as the corrected adaptive start control law.

[0041] Example 5 In practical implementation, after generating the corrected adaptive start-up control law through online rolling optimization, the central control unit immediately executes the corrected adaptive start-up control law to complete the one-click continuous control of the subsequent start-up phase. The central control unit maintains a complete sequence of control command frames for the corrected adaptive start-up control law in its internal control command buffer. Each control command frame in this sequence is marked as either "pending execution" or "executed." Control command frames marked as "executed" are those already executed by the test bench field controller in the current control cycle or earlier, while those marked as "pending execution" are subsequent control command frames that have not yet been issued. At the arrival of each control cycle, the central control unit retrieves the earliest timestamped control command frame marked as "pending execution" from the control command buffer, uses this control command frame as the current control command frame, changes its status to "executed," and issues it to the test bench field controller via the industrial Ethernet bus.

[0042] In practice, after receiving the current control command frame from the central control unit, the test bench field controller performs cyclic redundancy check and unpacking operations on the current control command frame, extracting the fuel flow rate setpoint and the adjustable guide vane angle setpoint contained in the current control command frame. It is important to note that the corrected adaptive start-up control law still contains all four control command components, but the online rolling optimization only corrects the fuel flow rate setpoint and the adjustable guide vane angle setpoint. Therefore, the ignition on / off state and starter generator torque setpoint in the current control command frame retain the original values ​​at the corresponding moments in the initial adaptive start-up control law. The test bench field controller writes the fuel flow rate setpoint from the current control command frame into the target value register of the proportional-integral-derivative controller in the inner loop of the fuel regulating valve position. The proportional-integral-derivative controller in the inner loop of the fuel regulating valve position drives the torque motor of the fuel regulating valve, causing the valve core of the fuel regulating valve to move to the opening position corresponding to the fuel flow rate setpoint, completing the fuel flow rate regulation operation. The test bench field controller writes the adjustable guide vane angle setpoint from the current control command frame into the target value register of the guide vane actuator position loop proportional-integral-derivative controller. The guide vane actuator position loop proportional-integral-derivative controller drives the electro-hydraulic servo valve and the guide vane actuator, and adjusts the angle of the adjustable guide vane to the position corresponding to the adjustable guide vane angle setpoint through the linkage mechanism, thus completing the guide vane angle adjustment operation.

[0043] In practice, after the test bench field controller completes the fuel flow adjustment and guide vane angle adjustment operations corresponding to the current control command frame, the test bench data acquisition system synchronously collects the latest feedback operating data of the engine according to a fixed control cycle. This latest feedback operating data includes five components: actual fuel flow, actual exhaust temperature, actual rotor speed, actual lubricating oil pressure, and actual vibration amplitude. The test bench data acquisition system packages the latest feedback operating data into feedback data frames and transmits them back to the central control unit via the industrial Ethernet bus. After receiving the latest feedback operating data, the central control unit extracts the actual rotor speed, actual exhaust temperature, and actual vibration amplitude from the latest feedback operating data as the current real-time engine status. The central control unit compares the actual rotor speed in the latest feedback operating data with the preset idle speed setpoint in the central control unit configuration parameter memory. The idle speed setpoint is the nominal value of the high-pressure rotor speed in the ground idle state corresponding to the engine type, which is 6800 rpm. The central control unit compares the actual exhaust temperature in the latest feedback operating data with the preset allowable idle temperature range in the central control unit's configuration parameter memory. The allowable idle temperature range is a closed interval from the lower limit of 420 degrees Celsius to the upper limit of 580 degrees Celsius for exhaust temperature in the idle state. Simultaneously, the central control unit determines whether the engine rotor speed has reached a preset proportion of the idle speed setpoint, which is set to 90%, or 6120 rpm. If the rotor speed has not yet reached the preset proportion of the idle speed setpoint, the central control unit determines that the engine is still in the acceleration start-up phase. At this time, the latest feedback operating data replaces the feedback operating data used in the previous control cycle, and the online rolling optimization process is re-executed, including calculating the speed deviation, temperature deviation, vibration margin, correcting the rolling time-domain optimization algorithm, and replacing the corrected future time period control command frame sequence. This updates the subsequent unexecuted control command frames in the corrected adaptive start-up control law. The above process is repeated in each control cycle. That is, in each control cycle, the central control unit completes the issuance of the current control command frame, the acquisition of the latest feedback operating data, the judgment of the idle state conditions, and the re-updation of the corrected adaptive start control law.

[0044] In practice, when the actual rotor speed in the latest feedback operating data within a certain control cycle reaches the preset ratio of the idle speed setpoint of 6120 rpm, the central control unit triggers the idle fuel flow switching logic. Before the idle fuel flow switching logic is triggered, the fuel flow setpoint in the corrected adaptive start control law is a real-time corrected value obtained by an online rolling optimization algorithm iteratively optimizing based on the speed deviation, temperature deviation, and vibration margin. After the idle fuel flow switching logic is triggered, the central control unit obtains the current ambient atmospheric pressure from the latest feedback operating data. This current ambient atmospheric pressure is one of the environmental parameters collected and continuously updated by the distributed sensor network of the test bench before startup. The central control unit uses the current ambient atmospheric pressure as input and queries the idle fuel flow atmospheric pressure correction curve pre-stored in the configuration parameter memory. This curve is a discrete data table calibrated by the engine manufacturer during type approval testing at various altitudes. Atmospheric pressure is the input variable, and the idle fuel flow setpoint is the output variable. The unit of atmospheric pressure is kilopascals, and the unit of the idle fuel flow setpoint is kilograms per hour. The range of atmospheric pressure values ​​in the data table covers the annual atmospheric pressure range that may occur at the test stand location. The central control unit performs linear interpolation on the idle fuel flow atmospheric pressure correction curve based on the current ambient atmospheric pressure to obtain the idle fuel flow setpoint corresponding to the current ambient atmospheric pressure. The central control unit simultaneously writes this idle fuel flow setpoint into the fuel flow setpoint components of all unexecuted control command frames in both the initial and corrected adaptive start control laws, overwriting the originally stored fuel flow setpoints in these control command frames. The ignition on / off state, starter / generator torque setpoint, and adjustable guide vane angle setpoint remain unchanged. Therefore, the fuel flow setpoints of all subsequent control command frames to be executed in the initial adaptive start control law and the modified adaptive start control law are uniformly replaced with the idle fuel flow setpoints corrected based on the current ambient atmospheric pressure.

[0045] In practice, after switching the idle fuel flow setpoint, the central control unit continues to execute the following operations in each control cycle: issuing the current control command frame, collecting the latest feedback operating data, and updating the corrected adaptive start control law. The difference from the control logic before the switch is that after the switch, the online rolling optimization algorithm no longer optimizes the fuel flow setpoint; the fuel flow setpoint remains unchanged at the idle fuel flow setpoint. The online rolling optimization algorithm only corrects the adjustable guide vane angle setpoint to maintain stable rotor speed and exhaust temperature. The central control unit continuously monitors the actual rotor speed and actual exhaust temperature in the latest feedback operating data within each control cycle. The central control unit determines that the engine has entered a stable idle state when both of the following conditions are met simultaneously: The first condition is that the actual rotor speed remains within a preset fluctuation range of the idle speed setpoint of 6800 rpm, which is set to ±50 rpm, meaning the actual rotor speed continuously falls within a closed interval of 6750 rpm to 6850 rpm; the second condition is that the actual exhaust temperature remains within a closed interval of the allowable idle temperature range of 420 degrees Celsius to 580 degrees Celsius. The central control unit starts a stabilization timer. When both conditions are met simultaneously, the stabilization timer begins to accumulate, incrementing by 20 milliseconds per control cycle; when either condition is not met, the stabilization timer is reset to zero and accumulation stops. The central control unit compares the current count value of the stabilization timer with a preset stabilization time threshold, which is set to 30 seconds. When the count value of the stabilization timer reaches the preset stabilization time threshold of 30,000 milliseconds, the central control unit determines that the actual rotor speed has remained within the preset fluctuation range of the idle speed setting value and the actual exhaust temperature has remained within the allowable range of the idle temperature for a duration exceeding the preset stabilization time threshold. At this time, the engine has stabilized in the idle state, and the one-button continuous control of the subsequent start-up phase is completed.

[0046] In practice, after completing the one-click continuous control of the subsequent startup phase, the central control unit displays the message "Startup complete, engine has entered idle state" on the human-machine interface. At the same time, it aligns the initial adaptive startup control law control command frames, the corrected adaptive startup control law control command frames, and the feedback operation data collected in each control cycle according to the timestamps of all control cycles during the entire startup process as a new startup record. Using the current engine model and the current test bench number as the primary key, the startup result identifier field is set to "success" and written to the historical startup record table in the local historical database of the test bench, providing historical startup success case data accumulation for subsequent tests.

[0047] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A one-button start adaptive control method for an aero-engine test stand, characterized in that, The method includes the following steps: In response to the one-click start command, the system collects the current environmental parameters of the test bench and the initial state parameters of the engine, and obtains a database of historical successful start cases. Based on the current environmental parameters and the engine initial state parameters, the historical successful start-up case database is subjected to multi-dimensional working condition clustering to determine multiple similar successful case clusters that match the current working condition; Based on the startup parameter sequence in the cluster of similar successful cases, a digital twin model is used to perform multi-objective prediction and correction of the startup process, and an initial adaptive startup control law is generated. The initial adaptive start control law is sent to the test bench field controller to execute the engine's first ignition and speed increase operation, and the engine's feedback operation data is collected in real time during the execution process; Based on the feedback operation data, the initial adaptive start-up control law is optimized online to obtain a corrected adaptive start-up control law, and one-click continuous control of the subsequent start-up stage is completed based on the corrected adaptive start-up control law.

2. The one-button start adaptive control method for an aero-engine test stand according to claim 1, characterized in that, In response to the one-button start command, the system collects current environmental parameters of the test bench and initial engine status parameters, and retrieves a database of historical successful start cases, including: The current ambient temperature, current atmospheric pressure, and current humidity, as well as the current lubricating oil temperature, current engine cylinder temperature, current fuel manifold pressure, and current rotor initial speed are synchronously collected through the distributed sensor network of the test bench, and used as the current environmental parameters and the engine initial state parameters. Using the engine model and test stand number as indexes, historical start records that match the current engine model and have a successful start result are retrieved from the local historical database of the test stand to obtain the historical start success case database. Each historical start record includes environmental parameters at the start time, engine initial state parameters, full start control command sequence, and full feedback state sequence.

3. The one-button start adaptive control method for an aero-engine test stand according to claim 2, characterized in that, Based on the current environmental parameters and the engine initial state parameters, the historical successful start-up case database is subjected to multi-dimensional operating condition clustering to determine multiple similar successful case clusters that match the current operating condition, including: The current environmental parameters and the engine initial state parameters are combined into a current operating condition feature vector; For each historical start-up record in the historical start-up success case database, extract the environmental parameters at the start-up time and the engine initial state parameters from that historical start-up record, and merge them into the historical operating condition feature vector of that historical start-up record; Calculate the weighted Euclidean distance between the current operating condition feature vector and the historical operating condition feature vector of each historical startup record. The weights of each dimension of the weighted Euclidean distance are determined by the sensitivity coefficients of each parameter to the startup success rate. Sort all historical startup records in ascending order according to the weighted Euclidean distance, and select a predetermined number of historical startup records at the top of the sort as a candidate record set; The historical startup records in the candidate record set are divided into three dimensions: the ambient temperature range at startup, the lubricating oil temperature range at startup, and the rotor initial speed range at startup, to obtain multiple clusters of similar successful cases. Each cluster of similar successful cases corresponds to a set of interval combinations.

4. The one-button start adaptive control method for an aero-engine test stand according to claim 3, characterized in that, The generation of the initial adaptive start-up control law includes: Extract the complete startup control command sequence of all historical startup records within each similar successful case cluster, and statistically average the control command values ​​at each time point to obtain the cluster average control command sequence corresponding to each similar successful case cluster. Extract the full feedback state sequence of all historical startup records within each similar success case cluster, and statistically average the feedback state value at each time point to obtain the cluster average feedback state sequence corresponding to each similar success case cluster. Using the current working condition feature vector as input, and the cluster average control command sequence of each of the multiple similar successful case clusters as the initial population, the startup process simulation is performed on the cluster average control command sequence corresponding to each similar successful case cluster using a digital twin model, so as to obtain the simulation feedback state sequence corresponding to each similar successful case cluster. Calculate the deviation sequence between the cluster average feedback state sequence of each similar successful case cluster and the simulation feedback state sequence corresponding to that similar successful case cluster, and perform prediction correction on the corresponding cluster average control command sequence based on the deviation sequence to obtain the corrected control command sequence corresponding to each similar successful case cluster. The multiple corrected control command sequences corresponding to the multiple similar successful case clusters are weighted and fused according to the sample size of each similar successful case cluster to generate the initial adaptive start-up control law.

5. The one-button start adaptive control method for an aero-engine test stand according to claim 4, characterized in that, The initial adaptive start-up control law is sent to the test bench controller to execute the engine's first ignition and speed increase operations, and during the execution, real-time feedback operating data of the engine is collected, including: The initial adaptive start control law is divided into multiple sequentially arranged control instruction frames according to a preset time rhythm. Each control instruction frame includes a fuel flow setpoint, an ignition on / off state, a starter generator torque setpoint, and an adjustable guide vane angle setpoint. The first control command frame is sent to the test bench field controller, which then controls the fuel regulating valve, ignition exciter, starter generator and guide vane actuator to move synchronously according to the first control command frame in order to perform the first ignition and speed increase operation. During the entire process of the engine speed accelerating from standstill to idle speed, subsequent control command frames are sequentially issued according to the preset time interval, and the engine feedback operation data is recorded in real time through the test bench data acquisition system. The feedback operation data includes actual fuel flow, actual exhaust temperature, actual rotor speed, actual lubricating oil pressure, and actual vibration amplitude.

6. The one-button start adaptive control method for an aero-engine test stand according to claim 5, characterized in that, Based on the feedback operation data, the initial adaptive startup control law is optimized online to obtain the revised adaptive startup control law, which includes: The actual rotor speed in the feedback operation data is compared with the expected rotor speed at the corresponding moment in the initial adaptive start-up control law to calculate the speed deviation value. The actual exhaust temperature in the feedback operation data is compared with the expected exhaust temperature at the corresponding moment in the initial adaptive start-up control law to calculate the temperature deviation value. The actual vibration amplitude in the feedback operation data is compared with the preset vibration safety threshold to calculate the vibration margin value; Based on the speed deviation value, the temperature deviation value, and the vibration margin value, the rolling time domain optimization algorithm is used to correct the control command frames that have not yet been executed in the initial adaptive start-up control law, so as to obtain the corrected future time period control command frame sequence. The modified adaptive startup control law is formed by replacing the original control command frames that have not yet been executed in the initial adaptive startup control law with the modified future time period control command frame sequence.

7. The one-button start adaptive control method for an aero-engine test stand according to claim 6, characterized in that, Based on the speed deviation value, the temperature deviation value, and the vibration margin value, the rolling time-domain optimization algorithm is used to correct the control command frames that have not yet been executed in the initial adaptive startup control law, including: Taking the current time as the optimization start time, and determining the optimization time domain end time with a preset optimization time domain length, the control instruction frames located between the optimization start time and the optimization time domain end time in the initial adaptive start control law are extracted as the instruction frame sequence to be optimized. The speed deviation value, temperature deviation value, and vibration margin value are used as state disturbance terms and superimposed on the fuel flow rate setpoint and adjustable guide vane angle setpoint in the command frame sequence to be optimized, respectively, to obtain the disturbed command frame sequence. With constraints that the engine rotor acceleration does not exceed a preset acceleration upper limit, the exhaust temperature does not exceed a preset temperature upper limit, and the vibration amplitude does not exceed the vibration safety threshold, and with the optimization objective of minimizing the difference between the expected rotor speed and the predicted rotor speed at the end of the optimization time domain, the fuel flow rate setpoint and the adjustable guide vane angle setpoint in the disturbed command frame sequence are iteratively optimized to obtain the corrected future time period control command frame sequence.

8. The one-button start adaptive control method for an aero-engine test stand according to claim 7, characterized in that, The one-button continuous control for the subsequent startup phase based on the modified adaptive startup control law includes: The first control command frame that has not yet been executed in the modified adaptive start control law is sent to the test bench field controller as the current control command frame, and the test bench field controller executes the fuel flow adjustment and guide vane angle adjustment operations corresponding to the current control command frame. After executing the current control command frame, the latest feedback operating data of the engine is collected, and the subsequent control command frames that have not yet been executed in the corrected adaptive start control law are updated again based on the latest feedback operating data. Repeat the issuance and update steps until the engine rotor speed reaches the idle speed set value and the exhaust temperature stabilizes within the allowable idle temperature range, thus completing the one-button continuous control of the subsequent start-up phase.

9. The one-button start adaptive control method for an aero-engine test stand according to claim 8, characterized in that, Repeat the issuance and update steps until the engine rotor speed reaches the idle speed set value and the exhaust temperature stabilizes within the allowable idle temperature range, including: The rotor speed of the engine is compared with the idle speed setpoint. When the rotor speed reaches a preset ratio of the idle speed setpoint, the fuel flow setpoint in the initial adaptive start control law and the modified adaptive start control law is switched to the idle fuel flow setpoint corrected based on the current ambient atmospheric pressure in the current environmental parameters. After switching to the idle fuel flow setpoint, the rotor speed and exhaust temperature are continuously monitored. When the rotor speed remains within the preset fluctuation range of the idle speed setpoint and the exhaust temperature remains within the allowable range of the idle temperature for a duration exceeding the preset stable time threshold, the one-button continuous control of the subsequent start-up phase is determined to be completed.

10. The one-button start adaptive control method for an aero-engine test stand according to claim 1, characterized in that, The digital twin model is constructed in the following way: a thermodynamic, aerodynamic and rotor dynamics simulation model is established based on the physical characteristic equation of the engine, and the key parameters in the simulation model are identified and calibrated using the full-process feedback state sequence in the historical successful start-up case database, so as to obtain a digital twin model that can perform high-precision simulation of the start-up process.

Images