An optimized assembly method for aircraft engine rotors

Abstract

The present invention relates to an optimized assembly method for an aero-engine rotor, comprising a sample learning phase of establishing a sample database through multiple groups of data; entering an application phase after obtaining the sample database, measuring the unbalance and geometric runout of a disc-shaft single-piece rotor to be assembled in advance and inputting them into a BP neural network system learned in the sample learning phase, and the BP neural network system outputs the relative assembly angle of the rotor disc-shaft single-piece to be assembled and the predicted value of the final assembly result of the rotor assembly; the operator judges whether the predicted value meets the actual technical requirements, and if not satisfied, changes the predicted value by adjusting or replacing the rotor disc-shaft single-piece until it is satisfied; assembling the rotor according to the satisfactory assembly result; measuring the unbalance and geometric runout of the assembled rotor assembly, and inputting them into the sample database of the BP neural network system. The method can predict the assembly result, realize assembly optimization by adjusting the parts status in advance, and avoid repeated assembly and adjustment.

Description

Technical Field

[0001] The invention belongs to the technical field of aeroengines, and in particular relates to an optimized assembly method for an aeroengine rotor. Background Art

[0002] The rotor of an aircraft engine is assembled from multiple disks. In order to obtain the best assembly performance of the rotor after assembly, each single disk needs to be measured and superimposed with projection values, namely geometric runout and imbalance, before assembly. The measurement results are marked on the rotor disk. By adjusting the relative assembly angles of each disk during the assembly process, the imbalance and geometric runout of each disk are offset, minimizing the geometric runout and imbalance of the assembly. When assembled in the conventional way, the unbalance of the rotor disk shaft is a vector measured on a balancing machine, and the geometric run-out of the rotor disk shaft is a vector measured on a superposition projection device. The two measurement items are independent of each other, and there is no correlation between the values ​​and the angles. Before assembly, the operator determines the relative assembly angles of each level of disks based on the principle and experience of vector complementarity. However, since geometric run-out and unbalance cannot often be taken into account at the same time, after the rotor disk shaft is assembled, the unbalance and geometric run-out of the rotor assembly often cannot be optimized at the same time. In addition, during the rotor assembly process, the assembly interface will be deformed due to the tightening force of the bolts and the extrusion of the stop fit. Therefore, even if only the geometric run-out or only the unbalance is considered, the geometric run-out and unbalance results of the rotor assembly after assembly are not the sum of the superimposed vectors of each piece.

[0003] For the above reasons, when assembling the engine rotor, the geometric runout and imbalance of the rotor after assembly often exceed the tolerance, resulting in unqualified assembly. Even if the assembly is qualified, it cannot be guaranteed that the assembly result is the optimal result. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides an aircraft engine rotor optimization assembly method, which inputs the superimposed projection value of the rotor disk shaft, the rotor unbalance, the relative assembly angle, and the final assembly result into a BP neural network system, and establishes the relationship between the superimposed projection value, the rotor unbalance, the relative assembly angle and the final assembly result through the BP neural network system. After the BP neural network has learned a large number of samples, it can output the relative assembly angle and the assembly result by inputting the superimposed projection value and the rotor unbalance, thereby achieving the function of optimizing the assembly.

[0005] An optimized assembly method for an aero-engine rotor comprises the following steps:

[0006] Step 1: Sample learning stage, a sample database is established by measuring multiple sets of data obtained from the imbalance and geometric runout of multiple high-pressure compressor rotors before and after assembly;

[0007] Step 2: After obtaining the sample database, enter the application stage, first measure the unbalance and geometric runout of the single-piece rotor of the disc shaft to be assembled;

[0008] Step 3: The geometric runout and unbalance of the rotor disc shaft measured in step 2 are input into the BP neural network system that has been learned in the sample learning stage. At this time, the test mode is entered, and the BP neural network system outputs the relative assembly angle of the rotor disc shaft to be assembled in step 2 and the final assembly result of the rotor assembly, that is, the predicted value of the geometric runout and unbalance of the rotor assembly; the operator judges whether the predicted value meets the actual technical requirements. If the operator is not satisfied with the result, that is, if it does not meet the requirements, the relative assembly angle and the predicted value of the assembly result of the rotor assembly are changed by adjusting or replacing the rotor disc shaft until a satisfactory assembly result is obtained;

[0009] Step 4: Assemble the rotor according to the relative assembly angle given by the satisfactory assembly result;

[0010] Step 5: Measure the imbalance and geometric runout of the rotor assembly assembled in step 4, and input the assembly result into the sample database of the BP neural network system as an assembly case for the system to continuously learn and correct.

[0011] The adjustment of the rotor disk shaft in step three includes mechanical processing repair and changing the assembly angle.

[0012] In the sample learning phase of step 1, the method for establishing a sample database specifically includes the following steps:

[0013] Step 1.1: Measure the unbalance of the rotor shaft on a balancing machine, measure the geometric runout of the rotor shaft on a superimposed projection device, and mark the measurement results on the surface of the rotor shaft with a marker;

[0014] Step 1.2: Assemble the high-pressure compressor rotor. The relative assembly angle of the rotor disk and shaft is determined by the operator based on the principle of vector complementarity or actual assembly experience, and recorded in detail.

[0015] Step 1.3: After assembly, measure the geometric runout of the rotor assembly on a superimposed projection device, and measure the unbalance of the rotor on a balancing machine to obtain the final assembly result;

[0016] Step 1.4: Input the rotor disk shaft single piece geometric runout and the rotor disk shaft single piece unbalance measured in step 1.1, the relative assembly angle recorded in step 1.2, and the final assembly result measured in step 1.3 as a set of data into the BP neural network system in the computer, wherein the rotor disk shaft single piece geometric runout and the rotor disk shaft single piece unbalance measured in step 1.1 and the relative assembly angle recorded in step 1.2 in the set of data are input values, and the final assembly result measured in step 1.3 is the output value;

[0017] Step 1.5: Repeat steps 1.1 to 1.3, measure and assemble the rotor multiple times, and input multiple sets of data into the BP neural network system. The BP neural network system completes the learning and obtains the sample database, and then enters the application stage.

[0018] In order to ensure the prediction accuracy in the prediction stage, the BP neural network system in step 1.5 needs to input at least 500 sets of data samples for learning and training, so that it can have ideal prediction accuracy and enter the application stage.

[0019] The beneficial effects of the present invention are as follows: by optimizing the assembly of rotors through the method provided by the present invention, the qualified rate of rotor assembly is increased by about 5%. Based on the assumption that 300 units are assembled each year and the assembly time of each engine is 26 hours, the working time saved each year is: 300 units × 26 hours / unit × 5% = 390 hours; the engine assembly quality is improved through the optimized assembly of rotors, and the qualified rate of test runs is increased by about 2%. Based on the assumption that 300 units are assembled each year and the test run cost of each engine is 300,000 yuan, the test run cost saved each year is: 300 units × 300,000 yuan / unit × 2% = 1.56 million yuan, which can bring great economic benefits.

[0020] This method can provide assembly plans, predict assembly results, and achieve assembly optimization by adjusting the parts status in advance, avoiding repeated assembly and adjustment, reducing the workload of operators, and optimizing the selection of rotor parts to achieve reasonable allocation of resources. At the same time, it can accumulate current assembly experience and avoid the situation where the rotor assembly quality is unstable due to human factors.

[0021] The method can be applied to determining the pre-assembly scheme of engine rotors of various types, and is particularly suitable for determining the pre-assembly scheme of repairing engine rotors. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A flow chart of the aircraft engine rotor optimization assembly method provided by the present invention; DETAILED DESCRIPTION

[0023] In order to better explain the present invention and facilitate understanding, the technical solutions and effects of the present invention are described in detail below through specific implementation modes in conjunction with the accompanying drawings.

[0024] An optimized assembly method for an aero-engine rotor, such as Figure 1 As shown, the specific steps include:

[0025] Step 1: Sample learning stage, a sample database is established by measuring multiple sets of data on the imbalance and geometric runout of multiple high-pressure compressor rotors before and after assembly.

[0026] In the sample learning phase of step 1, the method for establishing a sample database specifically includes the following steps:

[0027] Step 1.1: Measure the unbalance of the rotor disc shaft on a balancing machine, measure the geometric runout of the rotor disc shaft on a stacking projection device, and mark the measurement results on the surface of the rotor disc shaft with a marker; the stacking projection device used in this embodiment is Aerospect SPS Stack Prediction System (stacked projection device);

[0028] Step 1.2: Assemble the high-pressure compressor rotor. The relative assembly angle of the rotor disk shaft, i.e., the assembly angle of each rotor disk shaft relative to the previous part, is given by the operator based on the principle of vector complementarity or actual assembly experience, and the relative assembly angle of the rotor disk shaft is recorded in detail;

[0029] Step 1.3: After the assembly is completed, the geometric runout of the rotor assembly is measured on the superposition projection device, and the unbalance of the rotor is measured on the balancing machine to obtain the final assembly result;

[0030] Step 1.4: Input the rotor disk shaft single piece geometric runout and the rotor disk shaft single piece unbalance measured in step 1.1, the relative assembly angle recorded in step 1.2, and the final assembly result measured in step 1.3 as a set of data into the BP neural network system in the computer, wherein the rotor disk shaft single piece geometric runout and the rotor disk shaft single piece unbalance measured in step 1.1 and the relative assembly angle recorded in step 1.2 in the set of data are input values, and the final assembly result measured in step 1.3 is the output value. The core algorithm of the BP neural network system realizes data input and output through the interface reserved according to the conventional settings according to actual needs;

[0031] Step 1.5: Repeat steps 1.1 to 1.3, measure and assemble the rotor multiple times, and input multiple sets of data into the BP neural network system. In order to ensure the prediction accuracy in the prediction stage, the BP neural network system needs to input at least 500 sets of data samples for learning and training to have ideal prediction accuracy and enter the application stage. When the data accumulates to more than 500 sets, the BP neural network system ends the learning and obtains the sample database, that is, the BP neural network system obtains more than 500 sets of input values ​​and output values. The BP neural network system obtains the relationship between input values ​​and output values ​​through more than 500 sets of data, achieving ideal prediction accuracy, and then enters the application stage.

[0032] Step 2: After obtaining the sample database, enter the application stage. First, measure the imbalance and geometric runout of the single-piece rotor on the disc shaft to be assembled.

[0033] Step 3: Input the geometric runout and imbalance of the rotor disc shaft measured in step 2 into the BP neural network system that has been learned in the sample learning stage. At this time, the test mode is entered. The BP neural network system will output the relative assembly angle of the rotor disc shaft to be assembled in step 2 and the final assembly result of the rotor assembly, that is, the predicted value of the geometric runout and imbalance of the rotor assembly, according to the relationship between the input value and the output value in the multiple sets of data obtained in the sample learning stage; the operator judges whether the predicted value meets the actual technical requirements. If the operator is not satisfied with the result, that is, it does not meet the requirements, the relative assembly angle and the predicted value of the assembly result of the rotor assembly are changed by adjusting or replacing the rotor disc shaft, so that the assembly plan and assembly result can be predicted and optimized in advance until a satisfactory assembly result is obtained. The adjustment of the rotor disc shaft includes mechanical processing repair and changing the assembly angle.

[0034] Step 4: Assemble the rotor according to the relative assembly angle given by the satisfactory assembly result.

[0035] Step 5: Measure the imbalance and geometric runout of the rotor assembly assembled in step 4, and input the assembly result into the sample database of the BP neural network system as an assembly case for the system to continuously learn and correct.

[0036] The method can predict the assembly result of the rotor assembly in advance through the geometric runout, unbalance amount and relative assembly angle of the rotor disk shaft single piece. Since the assembly result can be predicted in advance, the rotor assembly result can be optimized by optimizing and adjusting the geometric runout, unbalance amount and relative assembly angle of the rotor disk shaft single piece, thereby improving the rotor assembly quality.

Claims

1. An aircraft engine rotor optimization assembly method, characterized in that: The specific steps include: Step 1: Sample learning stage, a sample database is established by measuring multiple sets of data obtained from the imbalance and geometric runout of multiple high-pressure compressor rotors before and after assembly; In the sample learning stage, the method for establishing a sample database specifically includes the following steps: Step 1.1: Measure the unbalance of the rotor shaft on a balancing machine, measure the geometric runout of the rotor shaft on a superimposed projection device, and mark the measurement results on the surface of the rotor shaft with a marker; Step 1.2: Assemble the high-pressure compressor rotor. The relative assembly angle of the rotor disk and shaft is determined by the operator based on the principle of vector complementarity or actual assembly experience, and recorded in detail. Step 1.3: After assembly, measure the geometric runout of the rotor assembly on a superimposed projection device, and measure the unbalance of the rotor on a balancing machine to obtain the final assembly result; Step 1.4: Input the rotor disk shaft single piece geometric runout and the rotor disk shaft single piece unbalance measured in step 1.1, the relative assembly angle recorded in step 1.2, and the final assembly result measured in step 1.3 as a set of data into the BP neural network system in the computer, wherein the rotor disk shaft single piece geometric runout and the rotor disk shaft single piece unbalance measured in step 1.1 and the relative assembly angle recorded in step 1.2 in the set of data are input values, and the final assembly result measured in step 1.3 is the output value; Step 1.5: Repeat steps 1.1 to 1.3, measure and assemble the rotor multiple times, and input multiple sets of data into the BP neural network system. The BP neural network system finishes learning and obtains a sample database, and then enters the application stage; Step 2: After obtaining the sample database, enter the application stage, first measure the unbalance and geometric runout of the single-piece rotor of the disc shaft to be assembled; Step 3: The geometric runout and unbalance of the rotor disc shaft measured in step 2 are input into the BP neural network system that has been learned in the sample learning stage. At this time, the test mode is entered, and the BP neural network system outputs the relative assembly angle of the rotor disc shaft to be assembled in step 2 and the final assembly result of the rotor assembly, that is, the predicted value of the geometric runout and unbalance of the rotor assembly; the operator judges whether the predicted value meets the actual technical requirements. If the operator is not satisfied with the result, that is, if it does not meet the requirements, the relative assembly angle and the predicted value of the assembly result of the rotor assembly are changed by adjusting or replacing the rotor disc shaft until a satisfactory assembly result is obtained; Step 4: Assemble the rotor according to the relative assembly angle given by the satisfactory assembly result; Step 5: Measure the imbalance and geometric runout of the rotor assembly assembled in step 4, and input the assembly result into the sample database of the BP neural network system as an assembly case for the system to continuously learn and correct.

2. The method for optimizing the assembly of an aircraft engine rotor according to claim 1, characterized in that: The adjustment of the rotor disk shaft in step three includes mechanical processing repair and changing the assembly angle.

3. The method for optimizing the assembly of an aircraft engine rotor according to claim 1, characterized in that: In order to ensure the prediction accuracy in the prediction stage, the BP neural network system in step 1.5 needs to input at least 500 sets of data samples for learning and training, so that it can have ideal prediction accuracy and enter the application stage.

Images