An APU health baseline construction method based on a no-label running state recognition

Abstract

This invention proposes a method for constructing an APU health baseline based on tagless operating status identification, belonging to the field of aircraft auxiliary power unit health diagnosis technology. This invention addresses the problem of difficulty in constructing a health baseline for aircraft auxiliary power units under tagless operating conditions, leading to misjudgments in health assessments. The steps of this invention include: S1: A13 message reception and preprocessing; S2: A13 parameter grouping; S3: potential operating status identification; S4: construction of a health reference sample set; S5: establishment of a state-aware health baseline model; S6: outputting the health baseline result.

Description

Technical Field

[0001] This invention relates to a method for constructing an APU health baseline based on tagless operating status identification, belonging to the field of aircraft auxiliary power unit health diagnosis technology. Background Technology

[0002] The aircraft auxiliary power unit (APU) provides ground power, bleed air, main engine starting, and environmental control support for civil aircraft. Its health status directly impacts aircraft release reliability and maintenance economics. Airlines can obtain on-wing operating parameters of the APU, such as exhaust temperature, bleed air pressure, guide vane opening, ambient temperature, and generator load, through ACARSA13 messages. However, A13 messages are essentially discrete operational snapshots, and parameters are simultaneously affected by environmental boundaries, bleed air load, electrical load, control actions, and component health status. Therefore, before using A13 messages to assess the health status of the APU, a health baseline model under comparable operating conditions needs to be established first.

[0003] Defects and shortcomings of existing technology: 1. Lack of explicit operating status labels. Existing A13 messages typically do not directly provide labels for gas supply, power supply, mixed load, or transitional status, which makes it easy for samples from different operating conditions to be mixed, affecting health baseline modeling.

[0004] 2. Univariate trends are prone to misinterpretation. An increase in exhaust temperature may stem from load changes, environmental changes, or control adjustments, and does not necessarily indicate health degradation. Directly using EGT trends or thresholds for judgment can easily confuse normal operating condition fluctuations with true health deviations.

[0005] 3. The health baseline lacks state awareness. Existing methods mostly use global models or anomaly detection models, which make it difficult to answer the question of "what kind of exhaust temperature response should a healthy APU exhibit under the current operating state". Summary of the Invention

[0006] To address the problem of difficulty in constructing a health baseline for aircraft auxiliary power units under unlabeled operating conditions, which leads to misjudgments in health assessments, this invention proposes a method for constructing an APU health baseline based on unlabeled operating status identification.

[0007] The technical solution adopted by the present invention to solve the above problems is as follows: The present invention includes the following steps: Step 1: Receive and preprocess the operating parameter messages of the aircraft auxiliary power unit to obtain a valid operating parameter sample set; Step 2: Based on the valid sample set of operating parameters, divide the operating parameters into operating status descriptors, health-sensitive responses, and auxiliary interpretation and verification variables; Step 3: Based on the running state descriptor, perform unlabeled clustering on the samples in the valid running parameter sample set to obtain the potential running state of each sample; Step 4: Based on the potential operating states, select healthy reference samples from the set of valid operating parameter samples to construct a set of healthy reference samples corresponding to each potential operating state; Step 5: Based on the health reference sample set, establish a health baseline model for each potential operating state, and obtain the mapping relationship between the operating state descriptor and the health exhaust temperature reference value; Step 6: Based on the potential operating status and health baseline model, output the healthy exhaust temperature reference value for the new operating parameter message.

[0008] Furthermore, the operating parameter messages in step 1 include the aircraft's A13 messages, field explanation files, and maintenance reliability records; The preprocessing of the runtime parameter message includes: unifying the fields in the runtime parameter message, and then performing missing value screening, outlier removal, time matching, lifecycle alignment, and normalization in sequence.

[0009] Furthermore, step 2 specifically includes: The operating parameters that characterize environmental conditions, load status, gas path pressure, and control status in the effective operating parameter sample set are determined as operating status descriptors; The exhaust temperature-related parameters used in the effective operating parameter sample set for health baseline modeling and residual calculation were identified as health-sensitive responses. The aircraft identification information, auxiliary power unit identification information, cumulative operating time, and replacement event records, which are collected in the effective operating parameter sample set for life cycle organization and post-validation, are identified as auxiliary explanatory validation variables.

[0010] Furthermore, step 3 specifically includes: Standardize the samples in the valid operating parameter sample set using only the running status descriptor; Unsupervised clustering is performed on the standardized operating state descriptors to group samples with similar environmental conditions, load states, gas path pressures and control characteristics into the same potential operating state. Based on the clustering results, obtain the potential running state number of each sample; The formula for dividing potential operating states is: (1); In formula (1), Let be the vector composed of all APU operating condition input sensor parameters at time t. Clustering function; The formula for determining the potential operating state number of a sample is: (2); In formula (2), For the first k The center of a potential operating state.

[0011] Furthermore, step 4 specifically includes: Based on the cumulative running time and replacement event records in the auxiliary interpretation verification variables, early service samples and samples far from the replacement window are screened from the effective operating parameter sample set. Based on operational stability indicators, stable operating samples are selected from early service samples and samples far from the replacement window; The stable operating samples are grouped according to their potential operating states, and a health reference sample set corresponding to each potential operating state is constructed.

[0012] Furthermore, step 5 specifically includes: For each potential operating state, extract the health reference sample set for that potential operating state; The regression model was trained using the running status descriptor as the input variable and the exhaust temperature in the health-sensitive response as the output variable. Based on the trained regression model, the mapping relationship between the operating status descriptor and the healthy exhaust temperature reference value is obtained.

[0013] The mapping relationship between the operating status descriptor and the healthy exhaust temperature reference value is as follows: (3); For the first k A health baseline model under various operating conditions. For healthy EGT reference values, As a state-aware health baseline, This refers to the operational status characteristics of the APU.

[0014] Furthermore, step 6 specifically includes: Preprocess and group the new operating parameter messages according to their functions, and obtain the operating status descriptor corresponding to the new operating parameter messages; Based on the running status descriptor, identify the potential running status to which the new running parameter message belongs; The health baseline model corresponding to the potential operating state is invoked, and the healthy exhaust temperature reference value is output according to the operating state descriptor. The healthy exhaust temperature reference value is used for residual calculation and health deviation estimation.

[0015] The beneficial effects of this invention are: (1) Continuous and sortable health indicators: This invention outputs continuous health deviation indicators, which can be used to sort health risks between different APUs and different life cycle stages.

[0016] (2) Improve comparability across operating conditions: This invention performs residual standardization within the same potential operating state, reducing misjudgments caused by different loads and environmental conditions.

[0017] (3) Enhanced multi-parameter health characterization capability: This invention not only utilizes EGT residuals, but also integrates bleed air, load, pressure and control variable deviations, which can more comprehensively reflect APU health changes.

[0018] (4) Support for maintenance window verification: This invention combines the health deviation indicators of the 30-day, 60-day and 90-day windows before replacement to make the results closer to the maintenance decision-making scenario. Attached Figure Description

[0019] Figure 1 A flowchart illustrating a method for constructing an APU health baseline based on label-free operational status identification; Figure 2 The image shows the recognition results for the unlabeled running state; Figure 3 A rendering of the health baseline model. Detailed Implementation

[0020] like Figure 1 As shown, the method for constructing an APU health baseline based on label-free operating status identification described in this embodiment includes the following steps: S1: A13 message reception and preprocessing; The system accesses A13 messages generated by the airline's ACARS system and combines them with maintenance reliability records to obtain lifecycle information such as aircraft number, APU serial number, installation time, replacement time, cumulative hours, and cumulative cycles. Then, based on the field interpretation file, it standardizes variable names and physical meanings, and completes missing value screening, outlier removal, timestamp correction, lifecycle matching, and variable normalization to obtain a valid A13 sample set.

[0021] S2: Grouping of A13 parameters; The preprocessed A13 parameters are divided into three categories: operating status descriptors, health-sensitive responses, and auxiliary interpretation and verification variables. Operating status descriptors include LCIT, PT, IGV, GLA, P2A, and WB, used to describe the environment, load, gas path pressure, and control status. Health-sensitive responses are represented by EGT, used for health baseline modeling. Auxiliary interpretation and verification variables include aircraft number, APU serial number, cumulative hours, cumulative cycles, and replacement events, used for lifecycle organization and post-hoc verification.

[0022] S3: Identification of potential operating states; Operating status descriptors such as LCIT, PT, IGV, GLA, P2A, and WB are standardized, and unsupervised clustering is used to identify potential operating states. During clustering, EGT, replacement events, or lifetime labels are not used, ensuring that A13 samples with similar environmental, load, gas path pressure, and control characteristics are grouped into the same potential operating state.

[0023] The potential running state is mapped as follows: (1); In formula (1), Let be the vector composed of all APU operating condition input sensor parameters at time t. Clustering function; The sample state is assigned as follows: (2); In formula (2), For the first k The center of a potential operating state.

[0024] S4: Construction of a healthy reference sample set; From all valid A13 samples, early service samples, samples far from the replacement window, stable operating samples, or high-quality samples are selected to form a health reference sample set. Replacement events are only used for sample selection and subsequent validation and are not used as input variables for the health baseline model.

[0025] S5: Establishment of a state-aware health baseline model; Within each potential operating state, a health baseline model is trained using health reference samples from that state to establish a mapping relationship between the operating state descriptor and the health EGT reference value. Gradient boosting regression, random forest, or extreme random tree regression models can be used in implementation. By modeling each state separately, health reference response estimation is completed under comparable operating conditions, avoiding baseline shift caused by mixing samples from different operating conditions.

[0026] The state-aware health baseline is: (3); For the first k A health baseline model under various operating conditions. For healthy EGT reference values, As a state-aware health baseline, This refers to the operational status characteristics of the APU.

[0027] S6: Outputs the health baseline results; For newly input A13 messages, preprocessing and parameter function grouping are first performed. Then, the potential operating state to which it belongs is identified, and the health baseline model for the corresponding state is invoked. The healthy EGT reference value of the sample under comparable operating states is output. The output results include the potential operating state number, the actual EGT value, the healthy EGT reference value, and the corresponding health baseline model number.

[0028] To verify the effectiveness of the state-aware health baseline established by this invention, the following simulation verification was conducted: This implementation method is verified based on the A13 message data from the on-wing APU of a certain airline. The results of potential operational status identification are as follows: Figure 2 As shown, samples of different states exhibit significant distribution differences in the LCIT, PT, IGV, GLA, P2A, and WB operating state descriptor spaces, indicating that this method can achieve comparable operating state classification without explicit operating condition labels. After establishing health baseline models based on different potential operating states, as shown... Figure 3 As shown, the model can output a health EGT reference value that matches the current operating conditions, indicating that the present invention has successfully established a state-aware health baseline for A13 messages.

[0029] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent substitutions, and improvements made to the above embodiments without departing from the scope of the present invention, based on the technical essence of the present invention and within the spirit and principles of the present invention, shall still fall within the protection scope of the present invention.

Claims

1. A method for constructing an APU health baseline based on label-free operating status identification, characterized in that, include: Step 1: Receive and preprocess the operating parameter messages of the aircraft auxiliary power unit to obtain a valid operating parameter sample set; Step 2: Based on the valid sample set of operating parameters, divide the operating parameters into operating status descriptors, health-sensitive responses, and auxiliary interpretation and verification variables; Step 3: Based on the running state descriptor, perform unlabeled clustering on the samples in the valid running parameter sample set to obtain the potential running state of each sample; Step 4: Based on the potential operating states, select healthy reference samples from the set of valid operating parameter samples to construct a set of healthy reference samples corresponding to each potential operating state; Step 5: Based on the health reference sample set, establish a health baseline model for each potential operating state, and obtain the mapping relationship between the operating state descriptor and the health exhaust temperature reference value; Step 6: Based on the potential operating status and the health baseline model, output the healthy exhaust temperature reference value for the new operating parameter message.

2. The method for constructing an APU health baseline based on tagless operating status identification according to claim 1, characterized in that, The operating parameter messages in step 1 include the aircraft's A13 messages, field explanation files, and maintenance reliability records; The preprocessing of the runtime parameter message includes: unifying the fields in the runtime parameter message, and then performing missing value screening, outlier removal, time matching, lifecycle alignment, and normalization in sequence.

3. The method for constructing an APU health baseline based on tagless operating status identification according to claim 1, characterized in that, Step 2 specifically includes: The operating parameters that characterize environmental conditions, load status, gas path pressure, and control status in the effective operating parameter sample set are determined as operating status descriptors; The exhaust temperature-related parameters used in the effective operating parameter sample set for health baseline modeling and residual calculation were identified as health-sensitive responses. The aircraft identification information, auxiliary power unit identification information, cumulative operating time, and replacement event records, which are collected in the effective operating parameter sample set for life cycle organization and post-validation, are identified as auxiliary explanatory validation variables.

4. The method for constructing an APU health baseline based on tagless operating status identification according to claim 1, characterized in that, Step 3 specifically includes: Standardize the samples in the valid operating parameter sample set using only the running status descriptor; Unsupervised clustering is performed on the standardized operating state descriptors to group samples with similar environmental conditions, load states, gas path pressures and control characteristics into the same potential operating state. Based on the clustering results, obtain the potential running state number of each sample; The formula for dividing potential operating states is: （1）； In formula (1), Let be the vector composed of all APU operating condition input sensor parameters at time t. For clustering functions, This represents the total number of operating condition categories; The formula for determining the potential operating state number of a sample is: （2）； In formula (2), For the first k The center of a potential operating state.

5. The method for constructing an APU health baseline based on tagless operating status identification according to claim 1, characterized in that, Step 4 specifically includes: Based on the cumulative running time and replacement event records in the auxiliary interpretation verification variables, early service samples and samples far from the replacement window are screened from the effective operating parameter sample set. Based on operational stability indicators, stable operating samples are selected from early service samples and samples far from the replacement window; The stable operating samples are grouped according to their potential operating states to construct the health reference sample set corresponding to each potential operating state.

6. The method for constructing an APU health baseline based on tagless operating status identification according to claim 1, characterized in that, Step 5 specifically includes: For each potential operating state, extract the health reference sample set for that potential operating state; The regression model was trained using the running status descriptor as the input variable and the exhaust temperature in the health-sensitive response as the output variable. Based on the trained regression model, the mapping relationship between the operating status descriptor and the healthy exhaust temperature reference value is obtained. The mapping relationship between the operating status descriptor and the healthy exhaust temperature reference value is as follows: （3）； In formula (3), For the first k A health baseline model under various operating conditions. For healthy EGT reference values, As a state-aware health baseline, This refers to the operational status characteristics of the APU.

7. The method for constructing an APU health baseline based on tagless operating status identification according to claim 1, characterized in that, Step 6 specifically includes: Preprocess and group the new operating parameter messages according to their functions, and obtain the operating status descriptor corresponding to the new operating parameter messages; Based on the running status descriptor, identify the potential running status to which the new running parameter message belongs; The health baseline model corresponding to the potential operating state is invoked, and a healthy exhaust temperature reference value is output according to the operating state descriptor. The healthy exhaust temperature reference value is used for residual calculation and health deviation estimation.

Images