Intelligent multi-modal data fusion control method and system
By constructing a transient intensity index for operating conditions and dynamically adjusting the fusion weights, the problems of frequent false alarms and low fault detection rate in multimodal data fusion of aero-engines were solved, and safe and precise control of the engine was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING CHANGKONG TECH CO LTD
- Filing Date
- 2026-06-17
- Publication Date
- 2026-07-14
AI Technical Summary
Existing technologies cannot dynamically adjust the fusion strategy according to the degree of transient changes in operating conditions in multimodal data fusion of aero-engines, resulting in frequent transient false alarms and low steady-state fault detection rate.
By constructing a transient intensity index for operating conditions, the fusion weights of vibration modes are dynamically adjusted. The high-frequency vibration and thermal distribution characteristic sequences are combined for weighted fusion, and a control correction quantity is generated. The amplitude of the control correction quantity is limited by the exponential function and the hyperbolic tangent function.
It achieves accurate monitoring of engine status under complex and variable operating conditions, balances the system's robustness to transient disturbances with its sensitivity to detecting real faults, reduces the false alarm rate, and improves the fault detection capability under steady-state conditions.
Smart Images
Figure CN122386628A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data fusion, and in particular to a control method and system for intelligent multimodal data fusion. Background Technology
[0002] As the core power plant of an aircraft, the real-time monitoring and precise control of the aircraft's operating status are crucial for ensuring flight safety. In actual flight missions or ground test benches, the Full Authority Digital Electronic Control System (FADEC) needs to process multimodal data from different physical channels, such as high-frequency vibration signals acquired through casing piezoelectric sensors and temperature field characteristics acquired through thermocouple arrays or infrared thermal imagers. These data exhibit complex nonlinear characteristics under different operating conditions, and how to effectively fuse this heterogeneous data to assist control decisions has always been a research hotspot in this field.
[0003] Existing technologies for processing the aforementioned multimodal data typically employ static weighted fusion or black-box fusion methods based on end-to-end neural networks. However, the operating conditions of aero-engines are extremely complex, encompassing steady-state conditions such as idle and cruise, as well as transient conditions such as afterburner and rapid deceleration. During transient changes in operating conditions, vibration amplitude naturally increases due to airflow excitation and mechanical clearance adjustments, which is a normal physical phenomenon. However, existing algorithms often use fixed thresholds or weights, making it difficult to distinguish between normal transient fluctuations and abnormal fault fluctuations. This leads to a high likelihood of misjudging vibration as exceeding limits during acceleration, triggering false alarms. Furthermore, traditional fusion strategies fail to dynamically adjust the focus based on physical mechanisms. For example, small distortions should be given high weight during steady-state cruise, while the impact of thermal inertial hysteresis data should be reduced during transient processes. Rigid fusion mechanisms cannot achieve a dynamic balance between suppressing false alarms and sensitive response.
[0004] To address the technical problem that existing technologies cannot dynamically adjust the fusion strategy of multimodal data according to the degree of transient changes in operating conditions of aero-engines, resulting in inaccurate control command generation or frequent false alarms, there is an urgent need for a control method that can sense physical operating conditions and adaptively adjust the fusion strategy. Summary of the Invention
[0005] To address the technical problem in existing technologies that cannot dynamically adjust the fusion strategy according to the degree of transient changes in operating conditions, resulting in frequent transient false alarms and low steady-state fault detection rates, this invention provides a control method and system for intelligent multimodal data fusion.
[0006] In a first aspect, the present invention provides a control method for intelligent multimodal data fusion, which adopts the following technical solution: A control method for intelligent multimodal data fusion includes the following steps: The system acquires high-frequency vibration sequences, thermal distribution characteristic sequences, and operating condition parameters during the operation of an aero-engine, and preprocesses and aligns the data. Based on the variation amplitude of core engine speed and throttle position command within the sampling period, an operating condition transient intensity index reflecting the stability of the engine's operating state is constructed. Dynamic fusion weights for vibration modes are determined based on the operating condition transient intensity index and the local variance of the high-frequency vibration sequences. These dynamic fusion weights decrease as the operating condition transient intensity index increases and maintain a response when the local variance increases significantly. The high-frequency vibration sequences and thermal distribution characteristic sequences are weighted and fused based on the dynamic fusion weights to obtain multimodal comprehensive characteristic values. These values are then combined with the control deviation between the target speed and the actual speed to generate the final control correction value, thereby adjusting the engine's operating state.
[0007] By constructing a transient intensity index reflecting the stability of the aero-engine's operating state and dynamically adjusting the fusion weights of vibration modes accordingly, the system overcomes the shortcomings of existing technologies that use fixed thresholds or static weights when processing multimodal data, which cannot distinguish between normal physical transient fluctuations and abnormal fault fluctuations. Specifically, when drastic changes in engine operating conditions lead to an increase in the transient intensity index, the system automatically reduces the weight of high-frequency vibration sequences, thereby effectively suppressing false alarms caused by normal physical phenomena such as airflow excitation. Meanwhile, it maintains a high response when local variance increases significantly or when the engine is in a steady state, thus achieving accurate monitoring of the engine's state under complex and variable operating conditions. This balances the system's robustness to transient disturbances with its sensitivity to detecting real faults, and generates a closed-loop control correction to regulate engine operation accordingly.
[0008] Preferably, the expression for the transient strength index under operating conditions is:
[0009] In the formula, for The transient intensity index of the operating conditions at any given time. It is a natural constant. Let be the variation of the core machine speed at time t within a unit sampling period. Let be the variation of the throttle position command at time t within a unit sampling period. and These are the sensitivity coefficients for changes in engine speed and throttle, respectively.
[0010] By introducing a mathematical expression that includes an exponential decay term with a natural constant, a transient intensity index is constructed to map the physical changes in engine speed and throttle position to a normalized numerical range. Utilizing the nonlinear characteristics of the exponential function, it can highly sensitively capture and quantify minute changes in the engine from steady state to transient state. Simultaneously, through a weighted combination of speed and throttle changes, it accurately reflects the engine's dynamic response characteristics with physical inertia, providing a precise and continuous quantitative benchmark for subsequent adaptive adjustment of the fusion weights.
[0011] Preferably, the method for obtaining the variation range of the core machine speed within a unit sampling period and the variation range of the throttle position command within a unit sampling period is as follows: calculate the absolute value of the difference between the core machine speed at the current moment and the previous moment to obtain the speed variation range, and calculate the absolute value of the difference between the throttle position command at the current moment and the previous moment to obtain the throttle variation range.
[0012] This method can capture instantaneous changes in control commands and system responses with minimal computational delay, ensuring the immediacy of the calculation of the transient intensity index of operating conditions.
[0013] Preferably, before determining the dynamic fusion weights of the vibration modal data, the method further includes: calculating a transient suppression factor for reducing the transient disturbance weights based on the transient intensity index of the operating condition, wherein the numerator of the factor is 1 and the denominator is 1 plus the sum of the transient intensity index of the operating condition.
[0014] By calculating the transient suppression factor, an explicit inverse relationship between the fusion weight and the transient intensity index of the operating condition was established. As the degree of engine transients intensifies, the value of this factor automatically decreases, thereby forcibly reducing the weight of data that are more affected by transient disturbances at the algorithm level. This fundamentally prevents the possibility of normal mechanical vibration increases during engine acceleration or deceleration being misjudged as faults, and reduces the false alarm rate in dynamic processes.
[0015] Preferably, before determining the dynamic fusion weights of the vibration modal data, the method further includes: calculating a signal saliency factor to characterize the intensity of signal fluctuations based on the local variance of the high-frequency vibration sequence, wherein the numerator of the factor is the local variance of the vibration signal within the current time sliding window, and the denominator is the local variance plus a preset positive constant.
[0016] By utilizing the ratio of the local variance within the sliding window to a preset constant, the fluctuation characteristics of the signal can be effectively extracted in a strong noise background. This ensures that even if the weight is transiently suppressed, if the vibration signal exhibits an anomaly far exceeding the normal fluctuation range, the factor can still maintain a certain weight contribution, preventing the underreporting of major safety faults due to excessive suppression and guaranteeing the safety of the system.
[0017] Preferred, dynamic fusion weights The expression is:
[0018] In the formula, The transient suppression factor at time t; The signal significance factor at time t; This is the steady-state gain coefficient. Let t be the transient intensity index of the operating conditions at time t.
[0019] By constructing a refined dynamic fusion weight calculation model, this model integrates transient suppression, signal saliency, and steady-state gain compensation. In particular, the introduction of a steady-state gain coefficient term enables the automatic amplification of the weight values when the transient intensity index is low, thereby improving the system's ability to capture minute distortions or early fault signals in steady state and solving the problem of traditional methods being insensitive to weak faults in steady state.
[0020] Preferably, the final control correction is generated by combining the control deviation between the target speed and the actual speed, satisfying the expression:
[0021] In the formula, Let be the final control correction amount at time t. This represents the gain coefficient of the PID controller. The control deviation between the target speed and the actual speed. For intelligent correction of gain coefficient, For multimodal integrated eigenvalues, It is the hyperbolic tangent activation function.
[0022] The multimodal integrated eigenvalues are activated by the hyperbolic tangent function and then incorporated into the final control correction calculation. Utilizing the saturation characteristics of the hyperbolic tangent function, it is ensured that regardless of fluctuations in the front-end fused eigenvalues, the resulting intelligent correction is always limited to the set safety gain range. This approach leverages the prior perception capabilities gained from multimodal data fusion for fine-tuning of PID control while preventing control command divergence due to data anomalies, thus strictly safeguarding the safety boundaries of the flight control system.
[0023] Preferred multimodal integrated eigenvalues The method for obtaining the multimodal comprehensive feature value is as follows: multiply the normalized vibration feature value with its corresponding dynamic fusion weight, multiply the normalized thermal feature value with its corresponding dynamic weight, and add the products of the two to obtain the multimodal comprehensive feature value.
[0024] By multiplying the normalized eigenvalues by their respective dynamic weights and then summing them, heterogeneous physical quantities are unified into a single comprehensive eigenvalue. This comprehensively utilizes the rapid response characteristics of mechanical vibration and the gradual change characteristics of thermal distribution, enabling control decisions to be based on more comprehensive physical state information and improving the accuracy of identifying complex coupled fault modes.
[0025] Preferably, the data is preprocessed and time-aligned, including: aligning the high-frequency vibration sequence, thermal distribution characteristic sequence and operating condition parameters with timestamps, and performing standardization using the sliding window method.
[0026] By employing rigorous timestamp alignment and sliding window standardization, the problems of inconsistent sampling frequencies and different dimensions among different sensors in aero-engines were resolved. This eliminated temporal deviations between data sources, removed mean drift and scale differences in the original data, and provided a unified, high-quality, and time-synchronized data foundation for the accurate fusion of multimodal data and the calculation of dynamic weights.
[0027] Secondly, the present invention provides an intelligent multimodal data fusion control system, which adopts the following technical solution: A control system for intelligent multimodal data fusion includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a control method for intelligent multimodal data fusion as described above is implemented.
[0028] The aforementioned intelligent multimodal data fusion control method is used to generate a computer program, which is then stored in a memory for loading and execution by a processor. This allows for the creation of a system based on the memory and processor, making it convenient to use.
[0029] The present invention has the following technical effects: By constructing a transient intensity index for operating conditions, the engine's operating state is quantitatively characterized, and a dynamic weighting mechanism is established accordingly: during transient acceleration, the vibration weight is automatically reduced to suppress false alarms caused by airflow disturbances; during steady state, the weight is increased through the gain coefficient to enhance the detection sensitivity for minor faults; and a bounded control correction quantity is generated by combining the hyperbolic tangent function. This solves the problem that existing technologies cannot simultaneously achieve transient anti-interference and steady-state high sensitivity, and realizes safe and precise control of the engine. Attached Figure Description
[0030] Figure 1 This is a flowchart of a control method for intelligent multimodal data fusion according to the present invention.
[0031] Figure 2 This is a schematic diagram comparing the system response under the same working conditions between the dynamic weight fusion strategy used in this embodiment of the invention and the fixed weight strategy used in the prior art. Detailed Implementation
[0032] 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, not all, of the embodiments of the present invention. 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.
[0033] This invention discloses a control method for intelligent multimodal data fusion, referring to... Figure 1 As shown, it includes the following steps: S101. Acquire high-frequency vibration sequences, thermal distribution characteristic sequences, and operating condition parameters during the operation of the aero-engine, and preprocess and time-align the data.
[0034] High-frequency vibration sequences are acquired using high-temperature piezoelectric accelerometers mounted on key sections of the engine casing (such as the fan casing and compressor casing). The sampling frequency is set above 10kHz, and bandpass filtering is applied to remove low-frequency rigid body mode interference and high-frequency electromagnetic noise. Thermal distribution feature sequences are acquired using an infrared thermal imager mounted on the side wall of the test bench. Specifically, for each frame of the infrared image, the highest temperature, average temperature, and variance of the temperature spatial gradient in key areas are extracted to form a low-dimensional numerical feature sequence. The original image pixels are not directly used as subsequent input to reduce computational latency. Core engine speed and throttle position commands are obtained from the engine control bus as operating condition parameters, with the sampling frequency synchronized with the control cycle, typically between 20Hz and 50Hz. The above data are timestamped and standardized using a sliding window method to construct the time series input vector to be processed.
[0035] Thus, by acquiring and preprocessing multi-source heterogeneous data, noise interference was removed and the data time reference was unified, providing a data foundation for subsequent feature extraction and fusion calculation.
[0036] S102. Based on the variation of core engine speed and throttle position command in the operating condition parameters within the sampling period, construct a transient intensity index reflecting the stability of engine operating status.
[0037] In order to accurately quantify whether the engine is currently operating smoothly or undergoing drastic changes, this invention utilizes the physical inertia of the speed response, which inevitably lags behind changes in throttle command, to construct a transient intensity index for operating conditions.
[0038] Specifically, the changes in engine speed and throttle are defined as follows:
[0039]
[0040] In the formula, This represents the change in core machine speed at time t within a unit sampling period. Let be the variation of the throttle position command at time t within a unit sampling period. , These are the core machine speeds at time t and t-1, respectively (unit: rpm). , These are the throttle position commands at time t and time t-1, respectively (unit: angle or percentage). This indicates taking the absolute value.
[0041] Furthermore, based on the aforementioned changes, a transient intensity index for the operating condition is constructed. :
[0042] In the formula, Let be the transient intensity index of the working condition at time t, and its value range is . , , Sensitivity coefficients for changes in engine speed and throttle (units are respectively) and These two coefficients serve as normalization factors, and their values are calibrated according to the engine model. They are used to map physical changes to the dimensionless exponential domain. , These represent the changes in core engine speed and throttle position command at time t within a unit sampling period.
[0043] For example, in a specific scenario, let's assume... .
[0044] Case 1 (Steady-state cruise): At any given moment, the core machine's rotational speed rpm, the previous moment rpm; throttle position command The previous moment .at this time Substitute into the formula, The index is close to 0, indicating that the operating conditions are extremely stable.
[0045] Case 2 (Rapid acceleration transient): At that moment, the accelerator was pushed hard, and the accelerator lever position command was... Mutation to ,Right now Due to mechanical inertia, the rotational speed only increases from... rpm increased to rpm, i.e. ;at this time ; Calculate the exponential term: ; The index rises and approaches 1, accurately representing the current drastic and transient state.
[0046] Thus, by constructing the transient intensity index of operating conditions, the severity of engine operating conditions can be quantified in real time and continuously, providing a quantitative basis for distinguishing between normal physical fluctuations and abnormal fault fluctuations.
[0047] S103. Based on the transient intensity index of the operating condition and the local variance of the high-frequency vibration sequence, determine the dynamic fusion weight of the vibration mode. The dynamic fusion weight decreases as the transient intensity index of the operating condition increases, and maintains its response when the local variance increases significantly.
[0048] To address the issue of varying reliability of modal data under different operating conditions, a physical-guided dynamic attention fusion mechanism is constructed. Taking vibration data as an example, the fusion weights are dynamically adjusted according to the transient intensity of the operating condition.
[0049] Specifically, for ease of calculation, the weight calculation is decomposed into transient suppression factors. and signal significance factor :
[0050]
[0051] This leads to the dynamic fusion weights of the vibration modes. :
[0052] In the formula, Let be the transient intensity index of the operating condition at time t. This represents the local variance of the vibration signal within the current time sliding window. For a preset positive constant (e.g.) ), used to prevent the denominator from being 0 and to set a basic threshold for the significance of the signal; This is the steady-state gain coefficient, dimensionless, used to compensate for vibration weights under steady-state conditions.
[0053] Following the example of step S102, set (Normalized variance threshold assumed to simplify calculations). .
[0054] In scenario two (rapid acceleration, normal airflow disturbance): Assume that due to the airflow disturbance, the local variance of the vibration signal naturally increases to... ,at this time , , Steady-state compensation term , .
[0055] Although the local variance of the vibration signal within the current time sliding window is large, the dynamic fusion weight is suppressed due to the large transient intensity index of the working condition, thus avoiding false alarms caused by excessive weight.
[0056] Compare to scenario three (steady-state cruise, minor malfunction): Return to the state of scenario one. Assuming a minor blade crack occurs at this point, the local variance of the vibration signal rises to [value missing]. , , The steady-state compensation term is , .
[0057] At this point, although the vibration amplitude is much smaller than that under rapid acceleration (0.05 < 0.5), it is still in a steady state. On the contrary, it is even higher (0.83>0.664), which reflects the system's high sensitivity to minor faults under steady state.
[0058] In this way, by calculating the dynamic fusion weight, the weight of vibration signals affected by airflow disturbances is automatically reduced during transient processes, thus suppressing false alarms; at the same time, the weight of minute abnormal signals is amplified by the gain coefficient in steady state, ensuring the sensitivity of fault detection.
[0059] S104. Based on dynamic fusion weights, the high-frequency vibration sequence and thermal distribution characteristic sequence are weighted and fused to calculate the multimodal comprehensive characteristic value, and combined with the speed control deviation to generate the final control correction quantity to adjust the engine operating state.
[0060] Calculate the multimodal integrated eigenvalues :
[0061] In the formula, The dynamic fusion weights of the vibration modes at time t are given. The vibration characteristic value at time t in the normalized high-frequency vibration sequence; The normalized thermal characteristic value at time t is specifically the weighted sum of the variances of the highest temperature, average temperature, and temperature spatial gradient of the key region in the infrared image at time t. The weights of the variances of the highest temperature, average temperature, and temperature spatial gradient are determined by the entropy weight method. The dynamic weights for thermal features are calculated using the same method as the dynamic fusion weights and can be configured based on thermal inertia characteristics.
[0062] Then calculate the final control correction amount. :
[0063] In the formula, The control deviation between the target speed and the actual speed; This represents the gain coefficient of the PID controller. This is the intelligent correction gain coefficient, used to limit the maximum correction authority of the multimodal fusion features on the control quantity; for example, its value is 5%. The hyperbolic tangent activation function maps the combined feature values to... Interval.
[0064] For example, if the calculated It is very large, for example, 2. The correction amount is close to ;like Extremely small, for example, 0.1. The correction amount is very small. This ensures that regardless of the fluctuations in the fused eigenvalues, the intervention in the control system is bounded, i.e., limited to... Within the specified range, safe fine-tuning based on traditional control laws was achieved.
[0065] The technical effects of this invention can also be illustrated in conjunction with the accompanying drawings. Figure 2 The invention demonstrates a comparison of health risk assessment values between the present invention and existing technologies. During the 20-30 second acceleration phase, the vibration amplitude naturally increases due to airflow disturbance. Existing technologies, with their fixed weights, incorrectly calculate risk values exceeding the alarm threshold, resulting in false alarms. In contrast, the present invention utilizes… The vibration weight was automatically reduced, keeping the risk value low and effectively suppressing false alarms. During the steady-state cruise phase after 85 seconds, when a simulated real fault occurred, due to this... At a low level, the present invention automatically restored high sensitivity, causing the risk value to rise rapidly and exceed the threshold, accurately triggering the alarm. This comparison intuitively demonstrates the superior performance of this solution in terms of steady-state sensitivity and transient robustness.
[0066] In this way, by fusing eigenvalues to generate bounded control corrections, multimodal data is used to finely adjust the control, while the correction magnitude is limited by the activation function, thus ensuring the safety and stability of the flight control system.
[0067] This invention also discloses an intelligent multimodal data fusion control system, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, a control method for intelligent multimodal data fusion according to the present invention is implemented.
[0068] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.
[0069] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.
Claims
1. A control method for intelligent multimodal data fusion, characterized in that, Including the following steps: The system acquires high-frequency vibration sequences, thermal distribution characteristic sequences, and operating condition parameters during the operation of an aero-engine, and preprocesses and aligns the data. Based on the variation amplitude of core engine speed and throttle position command within the sampling period, an operating condition transient intensity index reflecting the stability of the engine's operating state is constructed. Dynamic fusion weights for vibration modes are determined based on the operating condition transient intensity index and the local variance of the high-frequency vibration sequences. These dynamic fusion weights decrease as the operating condition transient intensity index increases and maintain a response when the local variance increases significantly. The high-frequency vibration sequences and thermal distribution characteristic sequences are weighted and fused based on the dynamic fusion weights to obtain multimodal comprehensive characteristic values. These values are then combined with the control deviation between the target speed and the actual speed to generate the final control correction value, thereby adjusting the engine's operating state.
2. The intelligent multimodal data fusion control method according to claim 1, characterized in that, The expression for the transient strength index under operating conditions is: In the formula, for The transient intensity index of the operating conditions at any given time. It is a natural constant. Let be the variation of the core machine speed at time t within a unit sampling period. Let be the variation of the throttle position command at time t within a unit sampling period. and These are the sensitivity coefficients for changes in engine speed and throttle, respectively.
3. The intelligent multimodal data fusion control method according to claim 2, characterized in that, The methods for obtaining the variation range of core machine speed and throttle position command within a unit sampling period are as follows: calculate the absolute value of the difference between the core machine speed at the current moment and the previous moment to obtain the speed variation range, and calculate the absolute value of the difference between the throttle position command at the current moment and the previous moment to obtain the throttle variation range.
4. The intelligent multimodal data fusion control method according to claim 1, characterized in that, Before determining the dynamic fusion weights of vibration modal data, the following steps are also taken: based on the transient intensity index of the operating condition, a transient suppression factor is calculated to reduce the weight of transient disturbances. The numerator of the transient suppression factor is 1, and the denominator is 1 plus the sum of the transient intensity index of the operating condition.
5. The intelligent multimodal data fusion control method according to claim 4, characterized in that, Before determining the dynamic fusion weights of vibration modal data, the following steps are also taken: based on the local variance of the high-frequency vibration sequence, calculate the signal significance factor used to characterize the intensity of signal fluctuations. The numerator of the factor is the local variance of the vibration signal within the current time sliding window, and the denominator is the local variance plus a preset positive constant.
6. The intelligent multimodal data fusion control method according to claim 5, characterized in that, Dynamic fusion weights The expression is: In the formula, The transient suppression factor at time t; The signal significance factor at time t; This is the steady-state gain coefficient. Let t be the transient intensity index of the operating conditions at time t.
7. The intelligent multimodal data fusion control method according to claim 1, characterized in that, Multimodal integrated eigenvalues The method for obtaining the multimodal comprehensive feature value is as follows: multiply the normalized vibration feature value with its corresponding dynamic fusion weight, multiply the normalized thermal feature value with its corresponding dynamic weight, and add the products of the two to obtain the multimodal comprehensive feature value.
8. The intelligent multimodal data fusion control method according to claim 7, characterized in that, The final control correction is generated by combining the control deviation between the target speed and the actual speed, satisfying the expression: In the formula, This represents the final control correction at time t. This represents the gain coefficient of the PID controller. The control deviation between the target speed and the actual speed. For intelligent correction of gain coefficient, For multimodal integrated eigenvalues, It is the hyperbolic tangent activation function.
9. The intelligent multimodal data fusion control method according to claim 1, characterized in that, Data preprocessing and time alignment include: aligning high-frequency vibration sequences, thermal distribution characteristic sequences, and operating condition parameters with timestamps, and standardizing them using the sliding window method.
10. A control system for intelligent multimodal data fusion, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a control method for intelligent multimodal data fusion according to any one of claims 1-9.