A method for diagnosing active suspension faults in new energy vehicles using fusion filtering and adaptive observers
By constructing a quarter-vehicle active suspension dynamics system model, integrating filters and adaptive observers, and designing a fault detection and isolation architecture, the problems of insufficient robustness of existing active suspension fault diagnosis methods in non-Gaussian noise environments and difficulty in identifying multiple types of faults are solved, achieving high-sensitivity and accurate fault detection and isolation.
Patent Information
- Application Number
- CN202611246598.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-17
- Publication Date
- 2026-11-17
AI Technical Summary
Existing active suspension fault diagnosis methods assume a Gaussian distribution when processing sensor measurement noise, which cannot adapt to non-Gaussian noise conditions, have poor noise disturbance resistance, and make it difficult to detect and isolate multiple types of complex faults simultaneously. The overly conservative adaptive threshold leads to insufficient diagnostic sensitivity.
A dynamic system model of a quarter-vehicle active suspension adapted to road disturbances, sensor noise, and various fault conditions is constructed. A filter and a fault detection observer are integrated to design a fault detection and isolation architecture. An adaptive parameter update law is used to generate dynamic thresholds. Noise and interference are suppressed by filters, and multiple sets of fault isolation estimators are constructed for accurate identification.
It significantly improves the robustness and fault detection sensitivity of the diagnostic system, accurately identifies actuator and suspension spring faults, provides reliable fault warnings and location information, and enhances the operational safety and reliability of new energy vehicles.
Smart Images

Figure CN123116570A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of chassis control and fault diagnosis technology for new energy vehicles, specifically relating to a fault diagnosis method for active suspension of new energy vehicles that integrates filtering and adaptive observers. Background Technology
[0002] With the booming development of the new energy vehicle industry, the active suspension system, as the core actuator of the intelligent chassis, is directly related to the vehicle's safety and ride quality in terms of operational reliability. Active suspension not only performs the basic function of connecting the vehicle body and wheels, but more importantly, it effectively isolates road bumps, significantly improving ride comfort. Therefore, it has long received widespread attention from academia and industry. Numerous research results have emerged regarding active suspension systems, covering active control algorithms, multi-performance objective trade-off design, actuator energy-saving control, and fault diagnosis and fault-tolerant control. Because the suspension system is constantly exposed to a complex external environment, its mechanical components and electronic components are prone to wear and aging. Actuators, sensors, and other key components may experience performance degradation or sudden failures, which will seriously weaken the ride comfort and operational reliability of the suspension system. Timely detection and accurate location of faults, followed by replacement or repair, can effectively prevent the chain reaction of faults and avoid major safety accidents. Therefore, research on fault diagnosis and fault-tolerant control of active suspension systems has significant theoretical and engineering value.
[0003] Extensive research has been conducted on fault diagnosis for active suspension systems. Under conditions of coexisting track irregularities and vehicle vibration noise, some researchers have proposed sensor fault detection methods suitable for rail vehicle suspension systems by extending the disturbances into system state variables and incorporating them into the model when designing the fault detection observer. For quarter-car active suspension systems, research has proposed fault diagnosis schemes that do not rely on empirical thresholds, introducing support vector machines as residual evaluators to intelligently evaluate the residuals generated by the fault detection observer, effectively avoiding the blindness of manual threshold selection. In suspension systems where external disturbances and sensor measurement noise coexist, fault detection and isolation methods based on Kalman filtering have been proposed. The core idea is to model external disturbances and measurement noise as zero-mean Gaussian white noise, and then use a filter bank to achieve fault detection and separation. Furthermore, for active suspension systems with parameter uncertainties and external disturbances, a patent proposes an adaptive fault detection and isolation method that effectively improves the sensitivity of fault diagnosis by updating uncertain parameters online when constructing the state estimator.
[0004] The above analysis reveals several common shortcomings in existing fault diagnosis methods that urgently need to be addressed. First, most methods simplify the handling of sensor measurement noise, often assuming it follows a zero-mean Gaussian distribution, which is inconsistent with the prevalent non-Gaussian, colored, or mixed noise distributions in real-world vehicle operating environments. This results in insufficient robustness of diagnostic algorithms under actual conditions. Second, existing research largely focuses on designing diagnostic solutions for single fault types (such as actuator or sensor failures), lacking a unified framework for detecting and isolating multiple fault types (such as the coexistence of suspension spring stiffness degradation and actuator failure). This makes it difficult to meet the diagnostic needs of complex fault scenarios in real-world systems. Furthermore, while existing adaptive observer methods can suppress parameter uncertainty to some extent, the dynamic thresholds they generate are often highly conservative, easily leading to missed or false alarms in noisy environments, thus requiring further improvement in detection sensitivity. Summary of the Invention
[0005] To address the technical problems in existing active suspension fault diagnosis methods, such as the simplistic and limited modeling of sensor measurement noise models, which fails to effectively adapt to non-Gaussian noise conditions, has poor noise disturbance resistance, makes it difficult to simultaneously detect and isolate multiple types of composite faults, and results in insufficient diagnostic sensitivity due to excessively conservative adaptive thresholds, this invention provides a new energy vehicle active suspension fault diagnosis method that integrates filtering and adaptive observers.
[0006] The new energy vehicle active suspension fault diagnosis method based on the fusion filtering and adaptive observer of the present invention includes the following steps:
[0007] S1. Construct a quarter-vehicle active suspension dynamics system model adapted to road disturbance, sensor noise, and multiple fault conditions, and calibrate the boundary constraints of the system's internal state variables, external bounded disturbances, sensor measurement noise, and the fault functions corresponding to the two types of suspension faults.
[0008] S2. Integrate the filter with the fault detection observer design and simultaneously build the fault detection architecture and the fault isolation architecture; the fault detection architecture outputs the filtered fault detection residual signal and the matched adaptive detection threshold, and the fault isolation architecture outputs each set of filtered fault isolation residual signals and the matched adaptive isolation threshold.
[0009] S3. First, compare the fault detection residual signal of the filtered form with the corresponding adaptive detection threshold to determine whether the suspension has failed. After determining that a fault has occurred, compare the fault isolation residual signal of each group of filtered forms with its corresponding adaptive isolation threshold to distinguish the fault type and complete the active suspension fault detection and fault isolation.
[0010] Preferably, step S1, which involves building a quarter-vehicle active suspension dynamics system model, specifically includes:
[0011] A quarter-vehicle active suspension dynamics system model is established, including sprung mass, unsprung mass, suspension springs, suspension damping, tires, and active suspension actuators. The vertical displacement of the vehicle body, vertical velocity of the vehicle body, vertical displacement of the wheels, and vertical velocity of the wheels are defined as system state variables. Bounded amplitude constraints are set for external road disturbances, unmodeled dynamics, sensor measurement noise, actuator fault parameters, and suspension spring fault parameters, and bounded amplitude constraints are also set for the rate of change of fault parameters. Two time intervals are divided: one where the fault has not occurred, and another where the fault has occurred. Fault functions corresponding to actuator faults and suspension spring faults are distinguished.
[0012] Preferably, the filter used in step S2 is a stable single-input single-output filter unit. This filter satisfies the strict positive real characteristic and all poles are located in the negative interval. It is used to uniformly filter and transform the state estimation error output by the adaptive observer.
[0013] Preferably, the design process of the fault detection architecture in step S2 is as follows:
[0014] A fault-free baseline fault detection observer is constructed, and a fixed observation gain is configured to estimate the vehicle body motion state in real time. The filter is used to perform filtering operations on the actual system state and the estimated state of the fault detection observer, respectively. The filtered state estimation error is used as the filtered fault detection residual signal, which is formed by the superposition of the filtering estimation error and the filtered sensor noise. Based on the boundary upper limits of various disturbances and noises in the system, an adaptive detection threshold that is dynamically updated over time is derived according to the error change law.
[0015] Preferably, the fault determination logic in step S3 is as follows: real-time acquisition of the filtered fault detection residual signal; when the instantaneous amplitude of the residual is greater than the synchronously updated adaptive detection threshold, the suspension is determined to have a fault, and the fault detection time is recorded.
[0016] Preferably, the fault isolation architecture in step S2 includes two sets of independent fault isolation estimators. The two sets of estimators correspond to the single fault scenarios of the actuator and the single fault scenario of the suspension spring, respectively. Each set of fault isolation estimators is configured with an independent observation gain and an adaptive update module.
[0017] Each fault isolation estimator has a fault function term matching its own fault type within its dynamic relationship, and the adaptive update module uses a parameter update law with projection operator constraints.
[0018] Preferably, each fault isolation estimator is equipped with an adaptive update rule for fault parameters with projection operator constraints; the update rule uses the state estimation error of this estimator and the known function of the corresponding fault type as driving signals; the update rule has a built-in numerical limit range to ensure that the fault estimation parameters obtained in real time fall within a preset reasonable range, while keeping the state estimation error and fault parameter estimation error bounded as a whole.
[0019] Preferably, the adaptive isolation threshold design method of the fault isolation architecture is as follows: calculate the dynamic adaptive isolation threshold separately for each group of fault isolation estimators; the threshold calculation incorporates multiple constraints such as the filter output signal, noise upper bound, external interference upper bound, adaptive estimation dynamic upper bound of fault parameters, and initial estimation error boundary.
[0020] Preferably, the fault isolation determination logic is as follows: after the fault detection time recorded in step S3, the filtered isolation residuals output by two sets of fault isolation estimators are continuously collected; if the output residual of a certain estimator is continuously lower than its corresponding adaptive isolation threshold, the suspension fault matched by that estimator is excluded; the fault corresponding to the estimator whose residual exceeds the corresponding adaptive isolation threshold is the current real fault type of the suspension.
[0021] Preferably, this method is equipped with a suspension active controller using a proportional-derivative control architecture; the controller outputs suspension active control force based on vehicle displacement and speed reference signals; the suspension active control force output by the controller and the suspension sensor measurement signals are collected synchronously throughout the process, serving as input data for the fault detection observer and the isolation estimator, so as to realize the synchronous and parallel operation of suspension active control and fault diagnosis.
[0022] The beneficial effects of this invention are:
[0023] This invention addresses the fault diagnosis needs of active suspension systems in new energy vehicles operating under complex environments by proposing a fault detection and isolation method that integrates filtering technology and an adaptive observer. The method first constructs an adaptive fault detection observer and a set of fault isolation estimators. It then utilizes an adaptive parameter update law to track system state changes in real time, generating isolation residual signals and dynamic isolation thresholds. Finally, a strictly positive and stable low-pass filter is introduced to perform collaborative frequency domain filtering on the residuals and thresholds. Based on the above technical solution, this invention achieves the following beneficial effects:
[0024] 1. This invention effectively suppresses non-Gaussian measurement noise and external interference, significantly improving the robustness of the diagnostic system.
[0025] Existing fault diagnosis methods typically assume sensor measurement noise to be zero-mean Gaussian white noise. This assumption is severely inconsistent with the non-Gaussian, colored, or mixed noise characteristics of the actual vehicle operating environment, leading to insufficient reliability of diagnostic algorithms under real-world conditions. This invention embeds filters with strictly positive real characteristics into the generation stages of the residual and threshold signals. Leveraging the filter's inherent ability to suppress high-frequency measurement noise and broadband external interference within a specific frequency band, it reduces the contamination of diagnostic results by noise and interference at the source. Simultaneously, because the filter is a bounded input with bounded output and stable, the filtered residual and threshold signals retain their bounded characteristics. Combined with the rigorous derivation of the dynamic threshold upper bound, this ensures that the adaptive threshold tightly encompasses the residual variation range under fault-free conditions. This mechanism effectively avoids overly conservative thresholds or false triggering caused by noise spikes or interference fluctuations, resulting in excellent robustness of the diagnostic system against non-Gaussian noise and unknown bounded interference.
[0026] 2. This invention significantly reduces the conservatism of the adaptive threshold and greatly improves the fault detection sensitivity.
[0027] In traditional observer fault diagnosis methods, to ensure that the residual does not exceed the threshold in fault-free conditions, the threshold is usually set relatively loosely. This conservative design is particularly prominent in noisy environments, directly leading to the missed detection of weak or early faults. This invention uses filtering technology to collaboratively process the residual signal and the threshold signal. The filtering operation not only acts on the residual to eliminate noise interference but also on various uncertainties in threshold generation, including the noise upper bound, the interference upper bound, and the initial error term. After filtering, the threshold upper bound is appropriately reduced, while the fault component in the residual signal is retained. At the time of fault occurrence, the filtered residual signal increases significantly due to the fault component, while the filtered threshold remains at a low level due to the suppression of noise and interference components, significantly increasing the margin between the two. Therefore, this invention can significantly reduce the triggerable threshold for fault detection without increasing the risk of false alarms, thereby achieving high-sensitivity detection of early and minor faults, effectively overcoming the inherent contradiction of traditional methods sacrificing sensitivity to ensure robustness.
[0028] 3. This invention constructs a multi-fault isolation estimator group, which realizes accurate identification of actuator faults and suspension spring faults.
[0029] Existing research mostly focuses on designing diagnostic solutions for single fault types, which is insufficient to meet the isolation requirements of complex fault scenarios in real-world systems. This invention constructs two sets of fault isolation estimators for two typical fault types: actuator faults and suspension spring faults. Each estimator is equipped with an independent adaptive parameter update law with a projection operator, ensuring that the estimated fault parameters remain within a bounded interval. When a fault of a certain type actually occurs, only the fault parameter estimates in the corresponding fault isolation estimator accurately track the true values, thus keeping the residuals of that estimator at a low level. Other mismatched fault isolation estimators, due to their model structures not matching the actual fault modes, will have significantly increased residuals exceeding their adaptive isolation thresholds. By using a decision logic that excludes the fault when the residual exceeds the limit, this invention can quickly identify the true fault type after fault detection, achieving accurate isolation between actuator and suspension spring faults, and providing clear fault location information for subsequent fault-tolerant control and precise maintenance.
[0030] 4. This invention combines theoretical rigor with engineering practicality.
[0031] This invention starts with a nonlinear quarter-vehicle active suspension dynamic model. Under reasonable conditions of bounded disturbances and bounded rates of change of fault parameters, it rigorously proves, using the Lyapunov method, that the estimation errors of the fault detection observer and the fault isolation estimator are globally consistent and bounded. Based on this, it derives analytical expressions for residuals and thresholds, as well as fault detection and isolation conditions, ensuring the reliability and reproducibility of the diagnostic method. Furthermore, this invention is designed for the actual operating conditions of active suspensions in new energy vehicles. It does not rely on the Gaussian assumption of noise and has no stringent requirements on sensor measurement accuracy and road excitation. The simulation verification uses mixed Gaussian noise and sinusoidal road input to fully simulate the real vehicle environment. Therefore, this invention can provide reliable fault warning, condition-based maintenance, and fault-tolerant control technology support for nonlinear vibration control systems such as active suspensions, intelligent chassis, and drive-by-wire chassis in new energy vehicles, which is of significant practical importance for improving the operational safety and reliability of new energy vehicles. Attached Figure Description
[0032] Figure 1 This is a schematic diagram of a quarter-axis active suspension model structure upon which the active suspension fault diagnosis method for new energy vehicles described in this invention is based.
[0033] Figure 2 This is a structural block diagram of the active suspension fault diagnosis method for new energy vehicles based on the fusion filtering and adaptive observer described in this invention.
[0034] Figure 3 This is a schematic diagram of the probability density function and time-domain signal of the mixed Gaussian noise used in the simulation verification of this invention, wherein... Figure 3 (a) is a schematic diagram of the probability density function. Figure 3 (b) is a schematic diagram of the time-domain signal;
[0035] Figure 4 This is a schematic diagram of the actuator fault detection results when the influence of sensor measurement noise is ignored in the fault detection method design of this invention;
[0036] Figure 5 This is a comparison chart of the detection results with and without filtering in the event of actuator failure, according to the present invention. Figure 5 (a) is a comparison chart of the detection residual and the detection threshold without filtering. Figure 5 (b) is a comparison chart of the detection residual and the detection threshold when filtering is applied;
[0037] Figure 6 This is a comparison chart of the detection residual and detection threshold of actuator faults when the present invention employs filtering processing;
[0038] Figure 7 This is a schematic diagram illustrating the isolation result of actuator faults when the influence of sensor measurement noise is ignored in the fault isolation method design of this invention. Figure 7 (a) is a comparison chart of the isolation residuals and adaptive isolation threshold of the first fault isolation estimator. Figure 7 (b) is a comparison chart of the isolation residuals and adaptive isolation thresholds of the second fault isolation estimator;
[0039] Figure 8 This is a schematic diagram illustrating the fault isolation result without filtering in the event of an actuator failure, according to the present invention. Figure 8 (a) is a comparison chart of the isolation residuals and adaptive isolation threshold of the first fault isolation estimator. Figure 8 (b) is a comparison chart of the isolation residuals and adaptive isolation thresholds of the second fault isolation estimator;
[0040] Figure 9 This is a schematic diagram illustrating the fault isolation results achieved by filtering in the event of an actuator failure, according to the present invention. Figure 9 (a) is a comparison chart of the isolation residuals and adaptive isolation threshold of the first fault isolation estimator. Figure 9 (b) is a comparison chart of the isolation residuals and adaptive isolation thresholds of the second fault isolation estimator;
[0041] Figure 10 This is a schematic diagram of the detection results of suspension spring faults when the influence of sensor measurement noise is ignored in the design of the fault detection method of this invention;
[0042] Figure 11 This is a comparison chart of the detection results of the present invention under the condition of suspension spring failure, with and without filtering. Figure 11(a) is a comparison chart of the detection residual and the detection threshold without filtering. Figure 11 (b) is a comparison chart of the detection residual and the detection threshold when filtering is applied;
[0043] Figure 12 This is a comparison chart of the detection residual and detection threshold of suspension spring faults when the filtering process is used in this invention.
[0044] Figure 13 This is a schematic diagram illustrating the isolation result of a suspension spring fault when the influence of sensor measurement noise is ignored in the fault isolation method design of this invention. Figure 13 (a) is a comparison chart of the isolation residuals and adaptive isolation threshold of the first fault isolation estimator. Figure 13 (b) is a comparison chart of the isolation residuals and adaptive isolation thresholds of the second fault isolation estimator;
[0045] Figure 14 This is a schematic diagram illustrating the fault isolation result without filtering in the event of a suspension spring failure, according to the present invention. Figure 14 (a) is a comparison chart of the isolation residuals and adaptive isolation threshold of the first fault isolation estimator. Figure 14 (b) is a comparison chart of the isolation residuals and adaptive isolation thresholds of the second fault isolation estimator;
[0046] Figure 15 This is a schematic diagram illustrating the fault isolation results achieved by filtering in the event of a suspension spring failure, according to the present invention. Figure 15 (a) is a comparison chart of the isolation residuals and adaptive isolation threshold of the first fault isolation estimator. Figure 15 (b) is a comparison chart of the isolation residuals of the second fault isolation estimator and the adaptive isolation threshold. Detailed Implementation
[0047] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Although some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the accompanying drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.
[0048] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.
[0049] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to"; the term "based on" means "at least partially based on"; and the term "one embodiment" means "at least one embodiment". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first," "second," etc., mentioned in this invention are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.
[0050] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".
[0051] The names of the messages or information exchanged between the multiple devices in the embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.
[0052] In the field of fault diagnosis for active suspension systems in new energy vehicles, the reliability of these systems, as the core actuators of intelligent chassis, directly impacts vehicle safety and ride comfort. During vehicle operation, the active suspension system relies on various sensors (such as vehicle displacement sensors and acceleration sensors) to perceive the vehicle's status in real time, thereby driving actuators to output active suspension control forces to suppress vehicle vibrations and isolate occupants from the impact of road unevenness. However, the suspension system is constantly exposed to a complex external environment, and the sensor signal acquisition and transmission process is inevitably affected by factors such as electromagnetic interference, temperature drift, and quantization errors, resulting in non-Gaussian, colored, or mixed noise components mixed in the measurement signals. Simultaneously, the randomness of road excitation, changes in vehicle load, and unmodeled dynamics also introduce bounded external disturbances. Under these conditions, key components such as actuators and suspension springs also face potential risks of wear and aging, potentially leading to failures such as reduced actuator output force or degraded spring stiffness. The combined effect of these factors makes fault diagnosis of active suspension systems a serious challenge.
[0053] From the technical principles of existing diagnostic methods, fault diagnosis methods based on analytical models typically use observers or Kalman filters to estimate the system state and detect faults by comparing the residuals between the actual measured values and the estimated values. The core premise of such methods is that the system model must be sufficiently accurate, and the statistical characteristics of noise and disturbances must be adequately described. However, in actual operating conditions, sensor measurement noise often does not satisfy the Gaussian distribution assumption of zero mean. This significantly reduces the accuracy of state estimation using traditional Kalman filters, and the residual signal contains significant noise components, leading to biases in subsequent fault determination. To address this issue, some studies have attempted to model external disturbances and measurement noise as an augmented state and perform joint estimation in the observer design. However, these methods are prone to numerical ill-conditioning as the model order increases, and it is difficult to guarantee estimation convergence under non-Gaussian noise conditions. Other studies have introduced adaptive observers to reduce the impact of model mismatch by estimating uncertain parameters online. However, the dynamic thresholds generated by such methods are often highly conservative. That is, the thresholds are set too wide under fault-free conditions to encompass all possible disturbances and noises. But when the fault amplitude is small, the residual signal is difficult to exceed the threshold, leading to missed faults.
[0054] To address the fault isolation problem, existing methods typically construct a set of isolation estimators corresponding to different fault types, determining the fault type by comparing the excesses of the residuals in each set with the corresponding thresholds. However, most studies only design isolation schemes for a single fault type (such as actuator or sensor faults), failing to adequately consider situations where multiple fault types coexist or similar fault characteristics are confused. In environments with strong measurement noise, the residual response pattern of one fault may overlap with the response pattern of another fault, leading to ambiguity in isolation decisions and further reducing the accuracy of fault location. Furthermore, the selection of thresholds in existing methods generally relies on experience or conservative estimates based on worst-case scenarios, lacking systematic and adaptive threshold design methods, making it difficult to balance sensitivity to weak faults and robustness to disturbances.
[0055] Based on the above analysis, the core technical defects of existing active suspension fault diagnosis methods can be summarized into three levels: First, the processing of sensor measurement noise is overly simplified, generally assuming it to be Gaussian white noise, which is seriously inconsistent with the non-Gaussian and colored noise characteristics in actual working conditions, resulting in insufficient robustness of the diagnostic method in real-world environments; Second, the overly conservative adaptive threshold makes it difficult to detect weak and early faults; Third, the detection and isolation of multiple types of faults lack a unified framework, resulting in insufficient positioning accuracy in scenarios with compound or similar faults.
[0056] To address the aforementioned technical problems, this invention provides a method for diagnosing active suspension faults in new energy vehicles that integrates filtering and adaptive observers. The core technical idea of this method is as follows: First, a dynamic model of a quarter-vehicle active suspension system incorporating multiple types of uncertainties is constructed to provide a precise mathematical benchmark for the design of subsequent diagnostic schemes. Based on this, a filter with specific frequency domain characteristics is co-designed with an adaptive observer. The filter is used to suppress measurement noise and external interference in the frequency domain, while adaptive techniques are employed to generate dynamic thresholds with adaptability to different operating conditions. Finally, by constructing an isolation architecture containing multiple sets of fault isolation estimators, accurate identification and location of actuator faults and suspension spring faults are achieved.
[0057] This invention provides a method for diagnosing active suspension faults in new energy vehicles by fusing filtering and adaptive observers, comprising the following steps:
[0058] S1. Construct a quarter-vehicle active suspension dynamics system model adapted to road disturbance, sensor noise, and multiple fault conditions, and calibrate the boundary constraints of the system's internal state variables, external bounded disturbances, sensor measurement noise, and the fault functions corresponding to the two types of suspension faults.
[0059] Specifically, the active suspension system of new energy vehicles faces multiple challenges under complex operating conditions. When a vehicle travels on a real road surface, road unevenness acts as an external excitation continuously on the suspension system, and its amplitude fluctuates with changes in road conditions. Simultaneously, sensors in the suspension system are inevitably affected by electromagnetic interference, temperature drift, and quantization errors during signal acquisition and transmission, resulting in noise components mixed in the measurement signals. In existing diagnostic methods, some researchers assume that sensor measurement noise is Gaussian white noise with a zero mean; however, non-Gaussian, colored, or mixed-distribution noise is prevalent in the actual vehicle operating environment. Furthermore, the suspension system is exposed to the external environment for extended periods, and its actuators and springs are at risk of wear and aging, potentially leading to failures such as reduced actuator output force or degraded spring stiffness. If the characteristic boundaries of these disturbances and failures are not accurately defined, the design of subsequent diagnostic schemes will face the risk of model mismatch, resulting in residual signals that cannot correctly distinguish between faults and interference, ultimately causing false alarms or missed alarms. Therefore, step S1 provides a precise mathematical model basis for the subsequent design of fault detection and isolation architecture by constructing a system model that includes the above-mentioned operating condition characteristics and calibrating the boundary constraints of various signals.
[0060] S2. Integrate the filter with the fault detection observer design and simultaneously build the fault detection architecture and the fault isolation architecture; the fault detection architecture outputs the filtered fault detection residual signal and the matched adaptive detection threshold, and the fault isolation architecture outputs each set of filtered fault isolation residual signals and the matched adaptive isolation threshold.
[0061] Specifically, the system model established in step S1 includes multiple uncertainties such as sensor measurement noise, external interference, and potential faults. If the original measurement signal is directly input into the observer for state estimation, noise and interference components will couple into the residual signal, causing the residual amplitude to exceed the fixed threshold even under fault-free conditions, leading to false alarms. Simultaneously, interference components will also increase the conservatism of the adaptive threshold, making it difficult to identify weak fault signals and causing missed alarms. To address these issues, step S2 introduces a filter into the diagnostic architecture design. This filter has specific frequency domain characteristics, capable of preserving fault characteristic frequency components while suppressing high-frequency components dominant in measurement noise and external disturbances. Therefore, after filtering the residual signal, the residual amplitude under fault-free conditions is effectively reduced, while the residual amplitude under fault conditions is highlighted. At the same time, the fault detection observer uses adaptive technology to generate a dynamic threshold, which can be adjusted online according to changes in the system's operating state, exhibiting stronger adaptability to operating conditions than a fixed threshold. Through the collaborative design of filters and fault detection observers, the fault detection architecture and fault isolation architecture can be built simultaneously. The former outputs the filtered fault detection residual signal and the matched adaptive detection threshold, while the latter outputs each set of filtered fault isolation residual signals and the matched adaptive isolation threshold.
[0062] S3. First, compare the fault detection residual signal of the filtered form with the corresponding adaptive detection threshold to determine whether the suspension has failed. After determining that a fault has occurred, compare the fault isolation residual signal of each group of filtered forms with its corresponding adaptive isolation threshold to distinguish the fault type and complete the active suspension fault detection and fault isolation.
[0063] Specifically, the filtered fault detection residual signal output by the fault detection architecture in step S2 reflects the deviation between the actual system state and the observer's estimated state. Under fault-free conditions, this residual is mainly caused by modeling errors, incompletely suppressed residual noise, and external interference, and its amplitude should be enveloped by the adaptive detection threshold. Once an actuator or spring fails, the system's dynamic characteristics change, and the deviation between the observer's estimated value and the actual value will increase significantly, causing the amplitude of the filtered residual to exceed the threshold envelope. Therefore, by comparing the relative magnitude of the residual and the threshold in real time, it can be determined whether a fault has occurred. After a fault is detected, the fault isolation architecture intervenes. The fault isolation architecture built in step S2 includes multiple sets of fault isolation estimators. Each set of estimators assumes that a specific fault type (such as actuator fault or suspension spring fault) is true and performs state estimation and adaptive parameter updates accordingly. If the actual fault type matches the fault type preset by a set of estimators, the residual of that estimator can remain at a low level; conversely, if the preset fault type does not match the actual fault, the residual will increase significantly and exceed the corresponding adaptive isolation threshold. Therefore, by comparing the exceedance of each group of isolation residuals with their corresponding thresholds, mismatched fault assumptions can be eliminated, the true fault type can be identified, and the entire process of fault detection and isolation can be completed.
[0064] Furthermore, step S1, which involves building a quarter-vehicle active suspension dynamics system model, specifically includes:
[0065] A quarter-vehicle active suspension dynamics system model is established, including sprung mass, unsprung mass, suspension springs, suspension damping, tires, and active suspension actuators. The vertical displacement of the vehicle body, vertical velocity of the vehicle body, vertical displacement of the wheels, and vertical velocity of the wheels are defined as system state variables. Bounded amplitude constraints are set for external road disturbances, unmodeled dynamics, sensor measurement noise, actuator fault parameters, and suspension spring fault parameters, and bounded amplitude constraints are also set for the rate of change of fault parameters. Two time intervals are divided: one where the fault has not occurred, and another where the fault has occurred. Fault functions corresponding to actuator faults and suspension spring faults are distinguished.
[0066] Specifically, the quarter-car suspension model simplifies the vehicle suspension system into two lumped mass units: sprung mass and unsprung mass. These are connected to the dampers via suspension springs. The unsprung mass contacts the road surface through the tires, and the actuators provide active suspension control forces. This model effectively reflects the vertical dynamic characteristics of the vehicle body and is a classic model widely used in active suspension control and fault diagnosis research. In this model, the system state variables are defined as the vehicle's vertical displacement, vehicle's vertical velocity, wheel's vertical displacement, and wheel's vertical velocity. Bounded amplitude constraints are set for external road disturbances, unmodeled dynamics, and sensor measurement noise, which aligns with the fundamental physical characteristic of finite disturbance energy in practical engineering. For actuator faults and suspension spring faults, this embodiment describes them as the product of fault parameters and known fault functions: the fault function for actuator faults is related to the actuator output force, and the fault function for suspension spring faults is related to the spring deformation. Furthermore, considering that the rate of change of fault parameters in practical engineering is constrained by the physical system's response capability and should not change infinitely quickly, bounded amplitude constraints are also set for the rate of change of fault parameters.
[0067] In a preferred embodiment, the nonlinear dynamic equations of the quarter-vehicle active suspension dynamics system model are expressed as follows:
[0068] (1)
[0069] (2)
[0070] In the formula, This refers to the sprung mass (the mass of the vehicle body, frame, and some inertial components). Unsprung mass (mass of suspension, wheel axles, and tires); This is the suspension spring stiffness coefficient; This refers to the tire stiffness coefficient. This refers to the suspension damping coefficient; This refers to the tire damping coefficient; The vertical displacement of the spring-loaded mass; This refers to the vertical displacement of the unsprung mass. Input vertical displacement for an unknown uneven road surface; For active suspension control force; The vertical acceleration of the sprung mass; The vertical acceleration of the unsprung mass; The vertical velocity of the sprung mass; The vertical velocity of the unsprung mass; This represents the rate of change of road surface displacement.
[0071] Define the system state vector ,in This refers to the vertical displacement of the vehicle body. The vertical speed of the vehicle body. This represents the vertical displacement of the wheel. The vertical speed of the wheel, indicated by the superscript. This represents the vector transpose operation. After considering potential faults, the suspension dynamics model can be rewritten in state-space form:
[0072] (3)
[0073] in, External interference and unmodeled dynamics in the vehicle body channel; External disturbances and unmodeled dynamics in the wheel channel; Number the fault type. This indicates an actuator malfunction. Corresponding to actuator failure, The basis functions are known for actuator faults; This indicates a suspension spring malfunction. Corresponding to suspension spring failure, The basis functions are known for the suspension spring fault; For vehicle body channel fault functions; For wheel path fault functions; For the first Unknown fault parameters; The upper bound of the fault parameter amplitude satisfies ; For fault segmentation time function; For continuous slow operation; The time when the fault occurred.
[0074] Fault segmentation time function expression:
[0075] (4)
[0076] When considering vehicle body sensor noise, the vehicle body sensor measurement model is expressed as follows: In the formula, Assign sensor channel numbers, For displacement channels, For speed channel, For the first The road sensor measures the output value; For the first The actual noise-free state of the road; For the first Road sensor measures noise.
[0077] When constructing the above system model, the following assumptions are given to support the design of subsequent fault diagnosis solutions:
[0078] Assumption 1: The perturbation is bounded and has positive constants. ,satisfy ; , These are the upper limits of the disturbance amplitude for the vehicle body and wheel channels, respectively.
[0079] Assumption 2: The rate of change of fault parameters is bounded, and normal values exist. ,satisfy ; , These are the time derivatives of the parameters for the first and second types of unknown faults, respectively. , These are the upper bounds of the change rates of the first and second type of unknown faults, respectively.
[0080] Furthermore, the filter used in step S2 is a stable single-input single-output filter unit. This filter satisfies the strict positive real characteristic and all poles are located in the negative interval. It is used to uniformly filter and transform the state estimation error output by the adaptive observer.
[0081] Specifically, the filter is a key component in this invention for achieving noise suppression and threshold compaction. In a fault diagnosis system, the state estimation error signal output by the observer contains a mixture of the actual estimation bias and noise components. If this error signal is directly used to construct the residual, the noise components will cause undesirable fluctuations in the residual, increasing the width of the threshold envelope and thus weakening the sensitivity of fault detection. The filter used in step S2 has strictly positive real characteristics and all poles are located in the negative interval, which means that the filter is stable and has bounded input-bounded output stability. When the input signal passes through this filter, its high-frequency noise components are effectively suppressed, while low-frequency fault characteristics are preserved. By simultaneously applying the filtering operation to the residual signal and the uncertainties in the threshold generation process, the residual and threshold are matched within the same frequency domain, avoiding the increase in conservatism caused by the overestimation of noise and interference components in the threshold, thus laying the foundation for improving the sensitivity of fault diagnosis.
[0082] In a preferred embodiment, the filter is a stable single-input single-output linear filter with a transfer function The format is as follows:
[0083] (5)
[0084] In the formula, The filter order; For the Laplace operator; and Design parameters for the filter. These are the coefficients of the filter numerator polynomial; The coefficients of the denominator polynomial of the filter; parameters , and The choice of transfer function makes All poles are negative real numbers, ensuring the filter is stable and has bounded input and bounded output characteristics.
[0085] For bounded signals ( ), For the first The upper bound of the noise amplitude measured by the road sensor is defined; the filtered signal is defined. , For the first The filtered noise signal after the road noise has passed through the filter; For filtering noise The upper bound of the amplitude satisfies .
[0086] To calculate the upper bound of the filter's output for a bounded input signal, the following upper bound mapping operator is introduced. For single-input single-output transfer functions and bounded signals ,definition for The corresponding bounded signal upper bound transfer operator is used to solve for the upper limit of the amplitude of the filtered output signal by using the upper bound of the input signal. The calculation formula is:
[0087] (6)
[0088] in, For transfer function The state-space implementation matrix, for 3D state matrix, for 3D input matrix, for The three output matrices satisfy the following conditions: ,and Corresponding impulse response function satisfy ; For matrix The standard type of Ruoerdang ; Let be an invertible matrix under the Jordan transformation; the dimension of the matrix is . ; for Unit matrix; Indicates taking the matrix The real part of each element. For a dimension of... any matrix , Defined as a matrix composed of the absolute values of its elements, where the subscripts are... The identifier for the element-level absolute value operation:
[0089] (7)
[0090] in, for OK An arbitrary matrix of dimension 1 For the matrix of the first Line 1 Column elements.
[0091] Based on the above upper bound mapping operator, the amplitude constraint of the filtered noise is as follows:
[0092] (8)
[0093] Furthermore, the design process of the fault detection architecture in step S2 is as follows:
[0094] A fault-free baseline fault detection observer is constructed, and a fixed observation gain is configured to estimate the vehicle body motion state in real time. The filter is used to perform filtering operations on the actual system state and the estimated state of the fault detection observer, respectively. The filtered state estimation error is used as the filtered fault detection residual signal, which is formed by the superposition of the filtering estimation error and the filtered sensor noise. Based on the boundary upper limits of various disturbances and noises in the system, an adaptive detection threshold that is dynamically updated over time is derived according to the error change law.
[0095] Specifically, based on the completed filter design, step S2 further constructs the fault detection architecture. First, a fault-free baseline fault detection observer is built. This observer assumes that the actuators and springs are functioning normally and is configured with a fixed observation gain to estimate the vehicle's motion state in real time. Under the premise that the observer is working normally, the difference between the actual system state and the observer's estimate originates only from initial estimation error, external interference, and sensor measurement noise. To suppress the influence of these factors on the residual, the actual system state and the observer's estimated state are filtered by the same filter. The difference between the two is the filtered fault detection residual signal. Due to the linearity of the filtering operation, this filtered residual can be decomposed into two parts: the filtering estimation error and the filtering sensor noise. The filtering estimation error originates from the observer's own estimation bias, while the filtering sensor noise originates from noise components in the measurement channel. Simultaneously, combining the upper limits of various system disturbances and noise, the evolution law of the filtering estimation error over time is derived using the dynamic analytical expression of the error, thereby constructing an adaptive detection threshold. This threshold can tightly encompass the range of variation of the filter residual under fault-free conditions, ensuring that the threshold is neither too conservative and masking faults, nor too aggressive and generating false alarms.
[0096] In a preferred embodiment, the dynamic equation of the fault detection observer is:
[0097] (9)
[0098] In the formula, This is an estimate of the vertical velocity of the sprung mass. and These are sensor measurements of the vertical displacement and velocity of the sprung mass, respectively. This is the observer gain.
[0099] Define filter variables:
[0100] (10)
[0101] In the formula, The signal is a filtered signal representing the actual vehicle speed. The filtered signal is used to estimate the vehicle speed.
[0102] Define the state estimation error after filtering Its time derivative is:
[0103] (11)
[0104] in, For the total composite disturbance term:
[0105] (12)
[0106] In the formula, The impulse response of the filter, i.e., the filter transfer function. The corresponding representation in the time domain; For state The initial value.
[0107] Filtered Fault Detection Residual Signal The calculation formula is:
[0108] (13)
[0109] In the formula, This is the sensor measurement of the vertical velocity of the sprung mass.
[0110] As can be seen from the above equation, the filtered residual signal is determined by the filtering estimation error. With filter sensor noise Formed by superposition.
[0111] Therefore, the filtering method is used to detect residual signals. The parsing expression is:
[0112] (14)
[0113] In the formula, For integration, the virtual time variable; This represents the initial filter estimation error.
[0114] The formula for calculating the adaptive detection threshold is:
[0115] (15)
[0116] in:
[0117] (16)
[0118] In the formula, An adaptive fault detection threshold; This is the upper bound of the initial filtering error. ; This is the upper bound function for the amplitude of the composite disturbance; For filter The corresponding bounded signal upper bound transfer operator is used to solve for the upper limit of the amplitude of the filtered output signal by using the upper bound of the input signal; The upper bound of the initial state value satisfies ; External interference The upper boundary.
[0119] Furthermore, the fault determination logic in step S3 is as follows: real-time acquisition of the filtered fault detection residual signal; when the instantaneous amplitude of the residual is greater than the synchronously updated adaptive detection threshold, the suspension is determined to have a fault, and the fault detection time is recorded.
[0120] Specifically, based on the fault detection architecture and adaptive detection threshold derived in step S2, step S3 provides a specific method for determining whether a fault exists. Since the filtered fault detection residual signal is enveloped by the adaptive detection threshold under fault-free conditions, once the instantaneous amplitude of the residual exceeds the threshold range, it indicates that the actual behavior of the system deviates from the dynamic mode described by the fault-free reference observer. At this point, it can be determined that a fault has occurred in the suspension system, and the fault detection time is recorded. This provides a time reference for the subsequent fault isolation phase.
[0121] In a preferred embodiment, the fault presence / absence determination logic is as follows:
[0122] If in Always satisfied If so, a fault is determined to have occurred, and the fault detection time is... .
[0123] Fault detection capability analysis: For an active suspension system, under the two conditions of the fault detection observer in formula (9) and the adaptive threshold in formula (15), if the fault function satisfies the following conditions:
[0124] (17)
[0125] The fault is in It was successfully detected at all times.
[0126] Furthermore, the fault isolation architecture in step S2 includes two sets of independent fault isolation estimators. The two sets of estimators correspond to the single fault scenarios of the actuator and the single fault scenario of the suspension spring, respectively. Each set of fault isolation estimators is configured with an independent observation gain and an adaptive update module.
[0127] Each fault isolation estimator has a fault function term matching its own fault type within its dynamic relationship, and the adaptive update module uses a parameter update law with projection operator constraints.
[0128] Specifically, the fault isolation architecture is designed based on the premise that fault detection has determined that the system has experienced an anomaly. The fault isolation architecture in step S2 comprises two sets of fault isolation estimators, each modeling based on a specific fault type as a premise. Specifically, the first set of estimators assumes that only actuator failure occurs in the system, and its dynamic equations include fault function terms that match the actuator failure. The second set of estimators assumes that the system only experiences suspension spring failure, and its dynamic equations contain a failure function term that matches the suspension spring failure. Each estimator group is independently configured with observation gain and adaptive update modules, without affecting each other. The core function of the adaptive update module is to perform online estimation of the fault parameters assumed by the estimator, so that the observer model gradually approximates the actual system state.
[0129] The projection operator constraint in the adaptive update rule is a crucial mechanism for ensuring the convergence of the estimated values. Since actual fault parameters are limited by physical constraints and have a finite range of variation, the projection operator restricts the parameter estimates to a pre-defined bounded interval. Internally, to prevent parameter drift. Projection operator. Its specific function is: for any input value, if the value exceeds the preset range... If the value is within the interval, it is projected onto the interval boundary; if the value is within the interval, it remains unchanged. Through the above projection operation, it is ensured that the fault parameter estimates obtained in real time are always within a preset reasonable range.
[0130] In a preferred embodiment, when the first One fault ( Corresponding to actuator failure, When a suspension spring failure occurs, the system dynamics can be written as:
[0131] (18)
[0132] No. The dynamic equations of the fault isolation estimator are as follows:
[0133] (19)
[0134] In the formula, For the first The fault isolation estimator provides an estimate of the vehicle's vertical motion speed. for The derivative of For the first Group fault parameter estimates, For the first Independent observation gain of the group fault isolation estimator.
[0135] Define the isolation state estimation error Fault parameter estimation error .
[0136] Furthermore, each fault isolation estimator is equipped with an adaptive update rule for fault parameters with projection operator constraints. This update rule uses the state estimation error of this estimator and the known function of the corresponding fault type as driving signals. The update rule has a built-in numerical limit range to ensure that the fault estimation parameters obtained in real time fall within a preset reasonable range, while the state estimation error and fault parameter estimation error remain bounded as a whole.
[0137] Specifically, the adaptive update rule essentially involves adjusting the estimated fault parameters online. This makes the state estimate of the fault isolation estimator... As close as possible to the actual state of the system Therefore, the driving signal for the update rule is composed of the state estimation error and a known function corresponding to the fault type. The physical meaning of the signal direction is: if the current estimate is too small, causing the estimation error to increase positively, then the update rule is activated. Positive adjustment occurs, and negative adjustment occurs otherwise. To ensure the stability of the algorithm, a damping term is introduced into the adaptive law to constrain the estimated velocity, and a projection operator is used to... Limited to a preset numerical range Inside.
[0138] In a preferred embodiment, the first The adaptive update law for each fault isolation estimator is:
[0139] (20)
[0140] in:
[0141] (twenty one)
[0142] In the formula, Adjust the positive constants to adaptively update the step size; It is a positive number; For interval projection operators, Preset parameter range. Projection operator. Its function is: for any input value, if the value exceeds the preset range... If the value is within the interval, then it is projected onto the interval boundary; if the value is within the interval, then it is left unchanged, thus ensuring... Always maintain Within the range. To adaptively update the driving signal.
[0143] Furthermore, the adaptive isolation threshold design method of the fault isolation architecture is as follows: calculate the dynamic adaptive isolation threshold separately for each group of fault isolation estimators; the threshold calculation incorporates multiple constraints such as the filter output signal, noise upper bound, external interference upper bound, adaptive estimation dynamic upper bound of fault parameters, and initial estimation error boundary.
[0144] Specifically, the calculation principle of the adaptive isolation threshold is similar to that of the detection threshold, but it requires consideration of more complex factors. The adaptive isolation threshold needs to encompass not only the residual fluctuations under fault-free conditions, but also the additional bias caused by model mismatch when the actual fault type differs from the estimator's preset fault type. Therefore, the threshold calculation process needs to integrate multiple constraints: the filter output signal reflects the influence of filtering characteristics on the residuals; the noise and interference upper bounds reflect the disturbance levels between the measurement channel and the external environment; the adaptive estimation dynamic upper bound of the fault parameters reflects the maximum possible deviation when the assumed fault type does not match; and the initial estimation error boundary reflects the transient characteristics during the observer's startup phase. The combined summation of these constraints constitutes a compact and sufficient adaptive isolation threshold envelope.
[0145] In a preferred embodiment, the first Filtered state estimation error of each fault isolation estimator Defined as:
[0146] (twenty two)
[0147] in, This is the filtered signal for the estimated value of the fault isolation estimator.
[0148] Filtered Fault Isolation Residual Signal The calculation formula is:
[0149] (twenty three)
[0150] For the first Fault isolation residual signal in the form of group filtering.
[0151] The analytical expression for the filtered state estimation error is:
[0152] (twenty four)
[0153] in, The time of fault detection. To isolate the initial error of the filter at the time of fault detection. For the first Known fault functions corresponding to the fault class, and isolation of composite disturbance terms. The expression is:
[0154] (25)
[0155] Adaptive isolation threshold The calculation formula is:
[0156] (26)
[0157] in:
[0158] (27)
[0159] In the formula, For the first Group adaptive isolation threshold; This serves as the upper limit for the composite disturbance in the isolation channel. This is the upper bound of the initial estimation error of the fault isolation estimator at the fault detection time. This is the dynamic upper bound for the adaptive estimation of fault parameters.
[0160] Furthermore, the fault isolation determination logic is as follows: after the fault detection time recorded in step S3, the filtered isolation residuals output by the two sets of fault isolation estimators are continuously collected; if the output residual of a certain estimator is continuously lower than its corresponding adaptive isolation threshold, the suspension fault matched by that estimator is excluded; the fault corresponding to the estimator whose residual exceeds the corresponding adaptive isolation threshold is the current real fault type of the suspension.
[0161] Specifically, at the time of fault detection Subsequently, the fault isolation module began continuously monitoring the filtered residual signals of the two sets of fault isolation estimators. If the... If the fault type preset by the estimator matches the actual fault, then the estimator's dynamic model matches the actual system structure, and its residuals are mainly affected by noise and interference, maintaining within the adaptive isolation threshold envelope. Conversely, if the preset fault type does not match the actual fault, then the estimator's model structure has a fundamental deviation (e.g., assuming an actuator failure but actually experiencing a suspension spring failure), even if the adaptive update law tries its best to adjust... Furthermore, it cannot completely bridge the residual increment caused by model mismatch, which will exceed the adaptive isolation threshold. Therefore, by identifying which set of estimators has residuals that are consistently below the threshold and which set of residuals exceeds the threshold, mismatched fault assumptions can be eliminated, and the true fault type can be accurately identified.
[0162] In a preferred embodiment, the fault isolation determination logic is as follows:
[0163] for If in At that moment, the The filter residuals of each fault isolation estimator satisfy ,in Then the first one can be excluded. The fault is determined to be the first fault. A fault occurs, and the fault isolation time is [time missing]. .
[0164] When the When the first fault occurs, the first ( A fault isolation estimator can be designed as follows:
[0165] (28)
[0166] in, For the first Each fault estimator gain.
[0167] Fault isolation capability analysis: When the first The fault is in When the fault occurs, the fault detection time is If time exists To quantitatively describe the fault isolation capability of the proposed method, a fault adaptation function is introduced here. It can describe the actual fault function. and its estimated value The difference between them, if the fault mismatch function satisfy:
[0168] (29)
[0169] in, Then the first The fault was isolated.
[0170] Furthermore, this method is equipped with a suspension active controller using a proportional-derivative control architecture; the controller outputs suspension active control force based on vehicle displacement and speed reference signals; the suspension active control force output by the controller and the suspension sensor measurement signals are collected synchronously throughout the process, serving as input data for the fault detection observer and the isolation estimator, thereby realizing the synchronous and parallel operation of suspension active control and fault diagnosis.
[0171] Specifically, the fault diagnosis method in this invention operates in conjunction with the active suspension controller. The controller employs a proportional-derivative (PDC) control architecture, calculating the active suspension control force output based on the deviation between the vehicle displacement reference signal and the measured vehicle displacement, the rate of change of vehicle displacement, and the measured vehicle speed. This active suspension control force... It is also an important input to the fault diagnosis observer and isolation estimator, because it is the fault function of the actuator fault. Directly related to the active suspension control force, the accurate acquisition of this force directly impacts the detection and isolation accuracy of actuator faults. Simultaneously, the suspension sensor measurement signals serve as feedback input to the observer, used for real-time updates of the state estimate. The control loop and diagnostic loop share the same control variables and measurement signals, operating in parallel without requiring additional modifications to the original control system, thus facilitating engineering implementation.
[0172] In a preferred embodiment, the expression for the proportional-derivative controller is:
[0173] (30)
[0174] in, This serves as a reference trajectory for the vehicle's motion displacement. The rate of change of the reference trajectory; and These are the proportional gain and derivative gain, respectively. The diagnostic module synchronously acquires the active suspension control force output by the controller. and suspension sensor measurement signals , This serves as the input data for the fault detection observer and the fault isolation estimator.
[0175] To verify the effectiveness of the method proposed in this invention, the following simulation verification examples are provided.
[0176] The reference values for the active suspension model of the new energy vehicle are shown in Table 1. In this simulation, the initial values of the state variables are set as follows: The observer's initial value is set to The initial value of the fault isolation estimator is set to , The initial value for fault parameter estimation is set to , The threshold parameter is set to... , , , The adaptive law parameters are set to... , , , Fault parameter constraints are set as follows: , The upper limit of the fault parameter is set to , The upper limit of noise is set to... , The upper limit of the filtered noise is set to , The observer gain is set to... , , The filter's transfer function is designed as follows: The suspension system is set to external interference. The upper limit of interference is set to A sinusoidal excitation signal is selected as the road surface input, and the road surface excitation signal is... The sensor measurement noise uses a mixture of Gaussian noise, whose probability density function and time-domain signal are as follows: Figure 3 As shown.
[0177] The gain of the proportional-derivative controller is set to , The vehicle displacement reference trajectory is set to .
[0178] Table 1 Parameter values of active suspension model for new energy vehicles
[0179]
[0180] To verify the effect of filtering on fault diagnosis performance in the proposed method, the following two comparative methods are used for simulation comparison:
[0181] Comparison Method 1: Completely ignore sensor measurement noise in the diagnostic method design, i.e., assume... , At this point, the filtering method used for fault detection results in the fault detection residual signal as follows: The filtering methods used for fault isolation are as follows: The residual signals for each fault isolation are... The corresponding adaptive detection threshold is... ,in External interference The upper bound of the threshold. The adaptive isolation threshold for each group is... ,in ( ) is a known function corresponding to the fault type. The dynamic upper bound for adaptive estimation of fault parameters is determined.
[0182] Comparison Method 2: Sensor measurement noise is considered in the diagnostic method design, but no filtering is used. In this case, the filtering form used for fault detection is the fault detection residual signal. The filtering methods used for fault isolation are as follows: The residual signals for each fault isolation are... The corresponding adaptive detection threshold is... ,in , and These are the upper bounds of the measurement noise for the displacement sensor and the velocity sensor, respectively. The corresponding adaptive isolation thresholds are... ,in .
[0183] The essential difference between the two comparison methods mentioned above and the method proposed in this invention lies in the fact that this invention uses the filter to filter the residual signal together with the uncertainties in the threshold generation process, and uses the upper bound transfer operator. Upper bound of the original noise and interference upper bound Compressed to the upper bound of filtered noise upper bound of interference after filtering This allows the adaptive detection threshold to tightly enclose the residual envelope when there is no fault and quickly expose the fault when there is a fault, effectively overcoming the shortcomings of comparison method 1, which ignores noise and causes false alarms, and comparison method 2, which considers noise but does not filter and causes missed alarms.
[0184] For actuator failure scenarios: In Inject actuator faults at all times, and set the fault parameters to... . Figure 4 The results of comparison method 1 (which ignores sensor measurement noise in the fault detection method design) are shown. It can be seen that the residual signal of the filtered fault detection still exceeds the adaptive detection threshold in the absence of faults, causing false alarms. This shows that if sensor measurement noise is not considered, the fault detection performance will be adversely affected. Figure 5 It gives in A comparison is made between the filtered fault detection residual signal and the adaptive detection threshold result when the actuator fails. Figure 5 (a) shows the result of comparison method 2 (without filtering). Figure 5(b) shows the result of the method of the present invention (with filtering). When the filter is used to process the fault detection residual signal and the adaptive detection threshold, the filtered fault detection residual signal can exceed the corresponding adaptive detection threshold after a fault occurs, effectively detecting the actuator fault. However, if no filtering is used (comparison method 2), the adaptive detection threshold is more conservative, and even if a fault occurs, the filtered fault detection residual signal never exceeds the adaptive detection threshold, resulting in a missed fault detection. The above results demonstrate that filtering is beneficial to improving fault detection performance. Figure 6 The results of actuator fault detection using the method of this invention are shown. At a certain moment, the residual signal of the filtered fault detection exceeds the corresponding adaptive detection threshold, thus achieving fault detection. Figure 7 The isolation results of Comparison Method 1 (ignoring sensor measurement noise in the fault isolation scheme) are shown. In the fault-free case, the filtered fault isolation residual signal of the first fault isolation estimator exceeds the adaptive isolation threshold, causing false alarms. Figure 8 and Figure 9 They were displayed in When an actuator failure occurs, the filter-based fault isolation residual signal and adaptive isolation threshold of Method 2 (without filtering) and the method of this invention (with filtering) are compared. Figure 9 It can be seen that after filtering, the filtered fault isolation residual signal of the second fault isolation estimator exceeds the corresponding adaptive isolation threshold, while the filtered fault isolation residual signal of the first fault isolation estimator remains below its adaptive isolation threshold, thus eliminating suspension spring faults and isolating actuator faults. In comparison, Figure 8 Without filtering, the filtered fault isolation residual signals of the two fault isolation estimators do not exceed their respective adaptive isolation thresholds, making it impossible to determine the fault type.
[0185] For suspension spring failure scenarios: In Constantly injecting suspension spring fault information, fault parameters set to... . Figure 10 The results of the comparison method 1 (ignoring sensor measurement noise in the fault detection design phase) are shown. Even when no fault occurs in the system, the residual signal of the filtered fault detection still exceeds the adaptive detection threshold, causing false fault alarms. Figure 11 The filtering method for suspension spring faults, the fault detection residual signal, and the adaptive detection threshold results are presented. Figure 11 (a) is comparison method 2 (without filtering). Figure 11(b) is the method of the present invention (with filtering). After applying filtering to transform the fault detection residual signal and the adaptive detection threshold, the filtered fault detection residual signal can exceed the adaptive detection threshold, successfully detecting the suspension spring fault; if filtering is not used, the adaptive detection threshold is too conservative, and the filtered fault detection residual signal can never exceed the adaptive detection threshold, resulting in missed fault detection. Figure 12 The results of suspension spring fault detection using the method of this invention are shown. At a certain time, the residual signal of the filtered fault detection exceeds the adaptive detection threshold, thus achieving effective fault detection. Figure 13 The results of the fault isolation method 1 (which ignores sensor measurement noise in the fault isolation design process) are shown. In the case of no fault, the filtered fault isolation residual signal of the first fault isolation estimator exceeds the adaptive isolation threshold, resulting in false alarms. Figure 14 and Figure 15 The results of filtering the fault isolation residual signal and adaptive isolation threshold for suspension spring failure are presented respectively. Figure 14 For comparison with method 2 (without filtering), Figure 15 This is the method of the present invention (with filtering processing). From Figure 15 It can be seen that, when using filtering, the filtered fault isolation residual signal of the first fault isolation estimator exceeds the corresponding adaptive isolation threshold, while the filtered fault isolation residual signal of the second fault isolation estimator never reaches the corresponding adaptive isolation threshold. Therefore, actuator faults can be ruled out, and the fault is determined to be a suspension spring fault. Figure 14 In the absence of filtering, the filtered fault isolation residual signals of the two fault isolation estimators do not exceed their respective adaptive isolation thresholds, making it impossible to clearly distinguish the fault type.
[0186] The simulation results fully verify that the method proposed in this invention can effectively achieve accurate detection and isolation of active suspension actuator faults and suspension spring faults in sensor measurement noise environments, and has higher detection sensitivity and isolation accuracy compared with existing methods.
[0187] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other described embodiments.
Claims
1. A method for diagnosing active suspension faults in new energy vehicles that integrates filtering and adaptive observers, characterized in that, Includes the following steps: S1. Construct a quarter-vehicle active suspension dynamics system model adapted to road disturbance, sensor noise, and multiple fault conditions, and calibrate the boundary constraints of the system's internal state variables, external bounded disturbances, sensor measurement noise, and the fault functions corresponding to the two types of suspension faults. S2. Integrate the design of filters and fault detection observers, and simultaneously build fault detection architecture and fault isolation architecture; The fault detection architecture outputs a filtered fault detection residual signal and a matched adaptive detection threshold, while the fault isolation architecture outputs a set of filtered fault isolation residual signals and a matched adaptive isolation threshold. S3. First, compare the fault detection residual signal of the filtered form with the corresponding adaptive detection threshold to determine whether the suspension has failed. After determining that a fault has occurred, the fault isolation residual signals of each group of filtered faults are compared with their corresponding adaptive isolation thresholds to distinguish the fault types and complete the active suspension fault detection and fault isolation.
2. The method for diagnosing active suspension faults in new energy vehicles using a fusion filter and adaptive observer as described in claim 1, characterized in that, Step S1, which involves building a quarter-vehicle active suspension dynamics system model, specifically includes: A quarter-vehicle active suspension dynamics system model is established, including sprung mass, unsprung mass, suspension springs, suspension damping, tires, and active suspension actuators. The vertical displacement of the vehicle body, vertical velocity of the vehicle body, vertical displacement of the wheels, and vertical velocity of the wheels are defined as system state variables. Bounded amplitude constraints are set for external road disturbances, unmodeled dynamics, sensor measurement noise, actuator fault parameters, and suspension spring fault parameters, and bounded amplitude constraints are also set for the rate of change of fault parameters. Two time intervals are divided: one where the fault has not occurred, and another where the fault has occurred. Fault functions corresponding to actuator faults and suspension spring faults are distinguished.
3. The method for diagnosing active suspension faults in new energy vehicles using a fusion filter and adaptive observer as described in claim 1, characterized in that, The filter used in step S2 is a stable single-input single-output filter unit. This filter satisfies the strict positive real characteristic and all poles are located in the negative interval. It is used to uniformly filter and transform the state estimation error of the adaptive observer output.
4. The method for diagnosing active suspension faults in new energy vehicles using a fusion filter and adaptive observer as described in claim 3, characterized in that, The design process of the fault detection architecture in step S2 is as follows: A fault-free baseline fault detection observer is built, and a fixed observation gain is configured to estimate the vehicle body motion state in real time. The filter is used to perform filtering operations on the actual system state and the state estimated by the fault detection observer, and the filtered state estimation error is used as the filtered fault detection residual signal. This residual is formed by the superposition of the filtering estimation error and the filtering sensor noise. By combining the upper limits of various disturbances and noises in the system, and based on the error change pattern, an adaptive detection threshold that is dynamically updated over time is derived.
5. The method for diagnosing active suspension faults in new energy vehicles using a fusion filter and adaptive observer as described in claim 4, characterized in that, The fault determination logic in step S3 is as follows: Real-time acquisition of the filtered fault detection residual signal; when the instantaneous amplitude of the residual is greater than the synchronously updated adaptive detection threshold, the suspension is determined to have a fault, and the fault detection time is recorded.
6. The method for diagnosing active suspension faults in new energy vehicles using a fusion filter and adaptive observer as described in claim 1, characterized in that, The fault isolation architecture in step S2 includes two independent fault isolation estimators. The two estimators correspond to the single fault scenarios of the actuator and the single fault scenario of the suspension spring, respectively. Each fault isolation estimator is configured with an independent observation gain and an adaptive update module. Each fault isolation estimator has a fault function term matching its own fault type within its dynamic relationship, and the adaptive update module uses a parameter update law with projection operator constraints.
7. The method for diagnosing active suspension faults in new energy vehicles using a fusion filter and adaptive observer as described in claim 6, characterized in that, Each fault isolation estimator is equipped with an adaptive update rule for fault parameters with projection operator constraints; the update rule uses the state estimation error of this estimator and the known function of the corresponding fault type as driving signals; The update rules have built-in numerical limit ranges to ensure that the fault estimation parameters obtained by real-time updates fall within a preset reasonable range, and at the same time can prove that the state estimation error and fault parameter estimation error remain bounded overall.
8. The method for diagnosing active suspension faults in new energy vehicles using a fusion filter and adaptive observer as described in claim 7, characterized in that, The adaptive isolation threshold design method of the fault isolation architecture is as follows: calculate the dynamic adaptive isolation threshold separately for each group of fault isolation estimators; the threshold calculation integrates multiple constraints such as the filter output signal, noise upper bound, external interference upper bound, adaptive estimation dynamic upper bound of fault parameters, and initial estimation error boundary.
9. The method for diagnosing active suspension faults in new energy vehicles using a fusion filter and adaptive observer as described in claim 8, characterized in that, The fault isolation determination logic is as follows: after the fault detection time recorded in step S3, the filtered isolation residuals of the outputs of the two sets of fault isolation estimators are continuously collected; if the output residual of a certain estimator is continuously lower than its corresponding adaptive isolation threshold, the suspension fault matched by that estimator is excluded; the fault corresponding to the estimator whose residual exceeds the corresponding adaptive isolation threshold is the current real fault type of the suspension.
10. The method for diagnosing active suspension faults in new energy vehicles using a fusion filter and adaptive observer according to any one of claims 1 to 9, characterized in that, This method is equipped with a suspension active controller using a proportional-derivative control architecture; the controller outputs suspension active control force based on vehicle displacement and speed reference signals; the active control force output by the controller and the measurement signals of the suspension sensors are collected synchronously throughout the process, and used as input data for the fault detection observer and the isolation estimator, so as to realize the synchronous and parallel operation of suspension active control and fault diagnosis.
