An intelligent diagnosis method and system for equipment based on mixed integer optimization

Abstract

The application discloses a kind of equipment intelligent diagnosis method and system based on mixed integer optimization, specifically relates to equipment intelligent diagnosis technical field, through the signal quality evaluation of each channel and linear programming weight self-adaption, automatic weight reduction of disturbed sensor is realized Stable fusion diagnosis under the interference environment, through mixed integer programming, introduce fault mutual exclusion, cause and coupling constraint, eliminate cross misjudgment, realize the accurate analysis of multi-system coupling fault, through the confidence four-level grading, and the safety key fault is uniformly adjusted alarm threshold, replace rigid binary output, realize the zero false alarm of high-risk fault, through the incremental tuning of real vehicle feedback data driving, automatically adapt vehicle model difference and vehicle long-term state change, shorten new model customization cycle.

Description

Technical Field

[0001] This invention relates to the field of intelligent equipment diagnostics, and more specifically, to an intelligent equipment diagnostics method and system based on hybrid integer optimization. Background Technology

[0002] In the current field of intelligent equipment diagnostics, diagnostic equipment needs to cover thousands of vehicle models and tens of thousands of complex faults. In actual repair scenarios, sensor signals are highly susceptible to noise such as high-frequency vibration and electromagnetic interference, leading to feature extraction deviations. This significantly increases the misjudgment rate of multi-system coupled faults (such as engine vibration caused by multiple systems such as ignition, fuel, and intake) that are already difficult to diagnose, often resulting in incorrect repairs by users or even secondary faults. At the same time, the sensor configurations of different vehicle models vary greatly. For each new model, engineers need to manually adjust the diagnostic logic and thresholds. This high-intensity manual customization not only has a long development cycle but also results in delayed model optimization, further amplifying the feature deviations caused by noise interference. Existing solutions can only make binary yes / no judgments when faced with suspected faults but insufficient evidence. Such binary judgments are prone to false negatives or false negatives under real vehicle signal fluctuations, especially in safety-critical scenarios such as braking systems, where the risk is unacceptable. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a device intelligent diagnostic method and system based on mixed integer optimization to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A device intelligent diagnostic method based on mixed integer optimization includes the following steps: Acquire the actual vehicle sensor signal stream, evaluate the signal quality channel by channel and extract fault features, and generate the initial confidence level of each sensor for candidate faults; Using the signal quality assessment results as constraints, a linear programming problem is solved to obtain the weighted fusion confidence level by solving the sensor fusion weights. The weighted fusion confidence score is input into a mixed integer programming model to solve for the optimal fault combination. Based on the confidence values ​​of each fault in the optimal fault combination, the diagnostic results are output in a graded manner. Collect real-vehicle diagnostic feedback, and incorporate confirmed, misjudged, and missed results into the training set to periodically trigger incremental tuning of diagnostic parameters.

[0005] In a preferred embodiment, generating the initial confidence level of each sensor for the candidate fault includes: The signal quality index is calculated for each sensor channel, and the time-domain and frequency-domain features are extracted to form a feature vector. The feature vector is compared with the standard feature vector in the pre-set fault template library to obtain the cosine similarity. The product of the signal quality index and the cosine similarity is used as the initial confidence level.

[0006] In a preferred embodiment, the linear programming of solving the sensor fusion weights to obtain the weighted fusion confidence includes: The maximum value of the initial confidence of each sensor channel for each candidate fault is taken as the fault sensitivity. A linear programming model is constructed with the goal of maximizing the sum of the products of the fusion weight of each channel, the signal quality index, and the fault sensitivity. The model satisfies the constraints that the sum of the weights is 1, the weights are not less than the product of the preset proportional coefficient and the signal quality index and are not higher than 0.5. The optimal fusion weight is obtained by solving the problem. For each candidate fault, the sum of the products of the optimal fusion weight of each channel, the signal quality index, and the initial confidence level is calculated as the weighted fusion confidence level.

[0007] In a preferred embodiment, the step of inputting the weighted fusion confidence into the mixed integer programming model to solve for the optimal fault combination includes: Obtain the weighted fusion confidence of candidate faults, the complexity coefficient of diagnostic rules, the list of mutual exclusion relationships and causal relationships between faults, as well as the effect coefficient, actual observed value and tolerance of each fault on the observed indicators. A mixed-integer programming model is constructed, with fault occurrence and rule activation as binary variables and severity as a continuous variable. The objective is to maximize the sum of the weighted fusion confidence and the product of the fault occurrence variable minus the cost of rule complexity and the cost of physical constraint violation. Mutual exclusion constraints, causal constraints, coupling constraints, severity correlation constraints and rule activation constraints are applied to solve for the optimal fault combination.

[0008] In a preferred embodiment, the complexity coefficient is the sum of the number of sensor channels on which the diagnostic rule depends and the number of judgment steps, and the effect coefficient is the ratio of the deviation of the observed index to the range of the observed index when the fault occurs alone.

[0009] In a preferred embodiment, the step of grading and outputting the diagnostic results includes: The confidence level of each fault in the optimal fault combination is compared with a preset threshold to classify the faults into four levels: confirmed faults, suspected faults, faults requiring attention, and faults not detected. Safety-critical faults are defined by system categories, and the threshold values ​​for each level of safety-critical faults are uniformly lowered by 0.15 compared to general faults.

[0010] In a preferred embodiment, the confirmed fault threshold for general faults is 0.85, the suspected fault threshold range is [0.70, 0.85), the threshold range for faults requiring attention is [0.50, 0.70), and the threshold for faults not detected is less than 0.50; the confirmed fault threshold for safety-critical faults is lowered to 0.70, the suspected fault threshold is lowered to 0.55, and the threshold for faults requiring attention is lowered to 0.35.

[0011] In a preferred embodiment, the collection of real-vehicle diagnostic feedback, including confirmation, misjudgment, and missed judgment results in the training set, and periodically triggering incremental tuning of diagnostic parameters, includes: Maintenance personnel mark and confirm each fault code output by the diagnostic system, identifying misjudged or missed faults. For missed faults, they trace back the feature vector and signal quality indicators at that moment and store them in the feedback database. Incremental tuning is triggered when the feedback database adds 50 new records or when 7 days have passed since the last tuning.

[0012] In a preferred embodiment, the incremental tuning includes: updating the standard feature vector of the fault template with the mean of the feature vectors of the confirmed records in the most recent time period; Adjust the lower bound ratio coefficient of the weight based on the number of false positives and false negatives caused by signal quality; Update the effect coefficient and tolerance using the mean and standard deviation of the observed index deviations in the confirmation records.

[0013] In a preferred embodiment, a device intelligent diagnostic system based on mixed integer optimization includes a signal confidence module, a weight fusion module, a coupling diagnostic module, a hierarchical output module, and a feedback optimization module. The signal confidence module is used to acquire the actual vehicle sensor signal stream, evaluate the signal quality channel by channel and extract fault features, and generate the initial confidence level of each sensor for candidate faults. The weighted fusion module is used to solve a linear programming problem for sensor fusion weights with signal quality assessment results as constraints, and obtain the weighted fusion confidence score. The coupled diagnostic module is used to input the weighted fusion confidence into the mixed integer programming model to solve for the optimal fault combination; The graded output module is used to grade and output the diagnostic results according to the confidence values ​​of each fault in the optimal fault combination. The feedback tuning module is used to collect real-vehicle diagnostic feedback, incorporate confirmed, misjudged, and missed results into the training set, and periodically trigger incremental tuning of diagnostic parameters.

[0014] The technical effects and advantages of this invention are as follows: 1. This invention addresses the problems of multi-fault coupling misjudgment under real vehicle conditions, unstable diagnosis due to sensor signal interference, long manual adaptation cycle for new models, and safety risks of missed reports due to binary judgment. It achieves stable fusion diagnosis under interference environment by automatically reducing the weight of disturbed sensors through channel-by-channel signal quality assessment and linear programming weight adaptation. It introduces fault mutual exclusion, causality and coupling constraints through mixed integer programming to eliminate cross-judgment and achieve accurate analysis of multi-system coupled faults. It replaces hard binary output by using a four-level confidence level and uniformly lowering the alarm threshold for safety-critical faults to achieve zero missed reports of high-risk faults. It automatically adapts to differences in vehicle models and long-term changes in vehicle status through incremental optimization driven by real vehicle feedback data, shortening the customization cycle of new models. Detailed Implementation

[0015] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Example 1: A device intelligent diagnostic method based on mixed integer optimization according to the present invention is given, including the following steps: Acquire the actual vehicle sensor signal stream, evaluate the signal quality channel by channel and extract fault features, and generate the initial confidence level of each sensor for candidate faults; Using the signal quality assessment results as constraints, a linear programming problem is solved to solve the sensor fusion weights, so that disturbed sensors are automatically deweighted and reliable sensors are replaced, thus obtaining the weighted fusion confidence level. The weighted fusion confidence is input into a mixed integer programming model to maximize the fault diagnosis confidence and minimize the rule complexity, and to solve for the optimal fault combination under the constraints of fault mutual exclusion, causality and coupling. Based on the confidence values ​​of each fault in the optimal fault combination, the diagnostic results are output in a graded manner, wherein the alarm threshold for safety-critical faults is lower than that for general faults. Collect real-vehicle diagnostic feedback, and incorporate confirmed, misjudged, and missed results into the training set. Periodically trigger incremental optimization of diagnostic parameters to adapt to differences in vehicle models and long-term changes in vehicle status.

[0017] In this embodiment of the invention, the signal stream of the actual vehicle sensors is acquired, the signal quality is evaluated channel by channel and fault features are extracted, and the initial confidence level of each sensor for candidate faults is generated, specifically including: The system acquires raw signal sequences from each sensor channel via the vehicle's CAN bus and performs the following for each channel: calculates signal quality metrics. ,in This is the ratio of the current signal power of the channel to the noise power under calibration conditions. This is the reference signal-to-noise ratio of the sensor under standard operating conditions. The standard deviation of the signal amplitude within the sliding window. The maximum permissible standard deviation of fluctuation for this sensor channel is set based on its physical characteristics. The physical rationality judgment value is V=1 when the signal value is within the sensor's measurement range and the rate of change between adjacent sampling points does not exceed the maximum physical rate of change calibrated by the sensor; otherwise, V=0. The weighting coefficient is... Furthermore, based on the sensor type, vibration sensors are preset to... , , Temperature sensors , , Pressure sensors , , ; Simultaneously, a data segment for the current analysis period is extracted from the signal sequence using a preset sliding window. Time-domain and frequency-domain features are extracted and concatenated to form the feature vector of the current observation for that channel. The time-domain features include the mean, peak value, variance of the signal amplitude within the window, and the number of times the signal crosses a preset reference level. The frequency-domain features are extracted by performing a Fast Fourier Transform on the data segment to obtain a discrete spectrum. Each frequency point in the spectrum corresponds to an amplitude, and the amplitude at the first harmonic of the engine rotation frequency is calculated. The amplitude corresponding to the second harmonic And the sum of the amplitudes of all frequency points within the fault-sensitive characteristic frequency band of the component monitored by the sensor. Simultaneously calculate the sum of amplitudes at all frequency points across the entire spectrum. Three ratios were obtained respectively. , , As the frequency domain feature; wherein, the engine rotation frequency is calculated from the engine speed signal synchronously acquired by the CAN bus. For example, when the current engine speed is 3000 rpm, the rotation frequency is 50 Hz. The first harmonic is taken as the spectral amplitude at 50 Hz, and the second harmonic is taken as the spectral amplitude at 100 Hz. Taking a vibration sensor for monitoring engine crankshaft bearing wear faults as an example, its fault-sensitive characteristic frequency band is pre-calibrated based on the passing frequencies of the bearing outer ring, inner ring, and rolling elements. For example, if the passing frequency of the bearing outer ring is 150 Hz, then the fault-sensitive characteristic frequency band of the sensor is taken as 145 Hz to 155 Hz, and the sum of the amplitudes at each frequency point within this frequency band is the frequency band. The feature vector is a fixed-length vector formed by concatenating the above time-domain features and three frequency-domain ratio features in a preset order, with each position in the vector corresponding to a fixed feature index. The feature vector is compared with each fault template in a pre-set fault template library. The fault template library contains several fault entries, and each fault entry stores the standard feature vector of each sensor channel under a known fault mode and the category identifier of the fault; for the first... The feature vector of the current observation of the channel and the first Calculate the cosine similarity between the standard feature vector of the channel in the fault template and the two. Specifically, the two vectors are multiplied dimension by dimension, summed, and then divided by the product of the magnitudes of the two vectors. The standard feature vector has the same dimension as the feature vector and the feature indices corresponding to each dimension are consistent. No. Channel to the first The initial confidence level of the candidate fault is set to 1. Therefore, when the signal quality deteriorates, the sensor's contribution to the determination of all candidate faults is automatically reduced.

[0018] In this embodiment of the invention, a linear programming problem is solved to obtain the sensor fusion weights, using the signal quality assessment result as a constraint. This problem automatically reduces the weight of disturbed sensors and replaces reliable sensors, resulting in a weighted fusion confidence score. Specifically, this includes: The first Sensor channel to the first Initial confidence of candidate faults As input, for each sensor channel The maximum initial confidence level of each candidate fault is taken as the fault sensitivity of that channel. Thus, each channel obtains a scalar. This indicates the sensor's ability to identify the most likely fault under the current signal characteristics; To determine the optimal fusion weights for each sensor channel under the current signal quality, a linear programming model is constructed: the decision variable is defined as the fusion weights of each sensor channel. ,common A continuous variable, For the total number of sensor channels, each The objective function is to maximize The product For comprehensive scoring, channels with high signal quality and high fault sensitivity receive higher comprehensive scores, and driving weight resources are preferentially allocated to these channels; constraints include weight normalization constraints. Weight lower bound constraint and upper bound constraints on weights ,in The preset lower bound ratio coefficient for weights is initially set to 0.05. This constraint links the minimum allowable weight for each channel to its real-time weight. The binding mechanism automatically shifts the minimum allowable weight downwards when signal quality deteriorates, while the upper bound constraint prevents any channel weight from exceeding 50%, thus avoiding over-reliance on a single sensor for diagnostics. Solving the linear programming model: The objective function coefficients, constraint matrix, and upper and lower bounds are input into the solver. The solution time limit is set to 10 milliseconds. The solver starts from a set of initial weights that satisfy all constraints and repeatedly adjusts the values ​​of each weight along the direction of increasing objective function value. After each adjustment, it checks whether the increment of objective function value has fallen below the preset minimum improvement amount. If it has fallen below the minimum improvement amount or reached the 10 millisecond time limit, the adjustment stops, and the weights at this time are output as the optimal fusion weights for each sensor channel. ; For the For each candidate fault, calculate its weighted fusion confidence. Therefore, channels with poor signal quality are due to and The contribution of the dual channel is reduced, while the channel with good signal quality and fault sensitivity automatically gets higher weight to complete the replacement. The confidence level after fusion has a stable characterization ability for real vehicle interference.

[0019] In this embodiment of the invention, the weighted fusion confidence score is input into a mixed integer programming model to maximize the fault diagnosis confidence score and minimize the rule complexity. The optimal fault combination is solved under fault mutual exclusion, causality, and coupling constraints, specifically including: Obtain each candidate fault Weighted fusion confidence , ,common Candidate faults are identified; simultaneously, the complexity coefficients corresponding to each rule are extracted from the diagnostic rule base. In addition, a list of mutually exclusive and causal relationships between faults is established, and the effect coefficients of each fault on various observed indicators are obtained from the vehicle physical model. and the actual values ​​of various current observation indicators With allowable tolerance ; Among them, the complexity coefficient The determination method is as follows: For each diagnostic rule, count the number of sensor channels that the rule depends on and the number of judgment steps within the rule, and add the two together as the result. For example, if a rule is completed based solely on the threshold of a single oxygen sensor signal, with one sensor and one judgment step, then... A certain rule requires timing matching of signals from a vibration sensor and a crankshaft position sensor, and involves three decision steps. If there are two sensors and three decision steps, then... Effect coefficient The determination method is as follows: for each type of fault The vehicle's operational status during isolated incidents was measured or simulated, and various observed indicators were recorded. The deviation relative to the fault-free state is taken as the ratio of this deviation to the range of the observed index. For example, when the fault "a certain cylinder's fuel injector is clogged" occurs alone, the exhaust temperature sensor reading rises by 15% relative to the normal value. Since the exhaust temperature sensor's range is -50℃ to 850℃, this rise represents 1.7% of the range. Therefore, this fault affects the observed index... ; Constructing a mixed-integer programming model: Defining binary decision variables , The diagnosis result indicates a fault. occur, Indicates that it will not occur; define a binary decision variable. , Indicates the first The diagnostic rule was activated in this diagnosis; define continuous decision variables. , indicating a fault The severity of the occurrence, 0 for no fault and 1 for the most severe condition; The objective function is to maximize ,in Take 0.02, We set the weights to 0.1 to control the weights of rule complexity cost and physical constraint violation cost, respectively. For the first The violation of a coupling constraint occurs when the predicted effect of a fault combination deviates from the observed value beyond the tolerance. Take the absolute value of the excess, otherwise return 0; The constraints include: Mutual exclusion constraint: For each pair of faults in a predefined set of mutually exclusive fault pairs ,Add to This prevents both from being selected simultaneously. Causal constraint: For each pair in the pre-defined set of causal fault relationships If the fault It will inevitably cause malfunctions Then add ; Coupling constraints: for each observation index ,Add to ,in, For the first The deviation between the current measured value and the normal baseline value of the observed index This constraint, which defines the allowable measurement tolerance for this indicator, requires that the overall effect of the selected fault combination matches the actual observed value. Severity-related constraints: Add Ensure the severity is 0 when there are no faults; add Ensure that the severity is at least 0.1 when a fault occurs; Rule activation association constraints: For each rule And the set of faults that it can determine; if any fault in this set is selected, then Set to 1, through constraints (For each fault in the set) )accomplish; Solving the mixed-integer programming model: The objective function coefficients, constraint matrix, variable types, and upper and lower bounds are input into the mixed-integer programming solver. The upper limit for the solution time is set to 500 milliseconds. The solver first extracts the most common fault combinations historically associated with the vehicle model from the vehicle knowledge base as an initial feasible solution, calculates the corresponding objective function value, and records it as the current optimal value. Then, it sequentially attempts to branch for each binary variable, fixing it to either 0 or 1. For each branch node, it solves the relaxed linear programming problem to estimate the optimal objective value that branch might reach. If the estimated value is lower than the current optimal value, the branch is abandoned and the process stops. If the estimated value is higher than the current optimal value, the branch continues until an integer feasible solution is obtained. Once an integer feasible solution is obtained, it is compared with the current optimal value and updated. The process stops when all branches have been explored or the solution time reaches 500 milliseconds, and the solution corresponding to the current optimal solution is output. Value and severity This constitutes the optimal fault combination.

[0020] In this embodiment of the invention, the diagnostic results are graded and output according to the confidence values ​​of each fault in the optimal fault combination, wherein the alarm threshold for safety-critical faults is lower than that for general faults, specifically including: Find the optimal fault combination, which includes all faults determined to have occurred. and their corresponding severity Simultaneously, the confidence scores of these faults are extracted from the weighted fusion confidence scores; for each fault in the optimal fault combination... , its confidence level Compared with a preset grading threshold, it is divided into one of four levels according to the following rules: Level 1, fault identification: when When the fault is confirmed, the fault code, fault name, and corresponding repair suggestions are output. The repair suggestions are obtained from the preset fault repair mapping table. The fault repair mapping table is indexed by the fault code. Each record contains the fault name, the location of the faulty component, the text of the repair operation steps, and a list of required tools. For example, the record corresponding to fault code P0301 is "Fault name: Cylinder 1 misfire, Location: Cylinder 1 spark plug and ignition coil, Repair steps: ① Remove cylinder 1 ignition coil ② Check spark plug electrode wear and carbon deposits ③ Replace spark plug or ignition coil, Required tools: spark plug socket, torque wrench"; Level 2, suspected fault: when If the fault is suspected, the fault code, fault name, and a list of recommended further tests are output. Recommended further tests are determined as follows: The fault is identified from the diagnostic rule base. For each of the associated detection steps, verify whether the sensor data required for each step meets the validation conditions in the current observation. The validation conditions are the sensor channel data. Furthermore, the cosine similarity between the feature vector corresponding to this step and the fault template is greater than or equal to 0.7. Detection steps that do not meet the verification conditions are selected as items for further detection. For example, the suspected fault "exhaust manifold leakage" is associated with three detection steps: checking the torque of the exhaust manifold connecting bolts, reading the voltage difference between the front and rear oxygen sensors, and performing a smoke test to check for leaks. If the current oxygen sensor channel... If the voltage difference is only 0.4, making the voltage difference data unreliable, then the output "It is recommended to conduct a smoke test to check for leaks" and "It is recommended to check the torque of the exhaust manifold connection bolts" will be used as further testing items. Level 3, attention should be paid to: when When the fault is identified as a state requiring attention, no fault code is output. Only the name of the abnormal indicator and the percentage of its current deviation from the reference value are output. The reference value is the value of various indicators of this model under the same working conditions without fault calibration. For example, the output "Engine exhaust back pressure is too high, currently deviating from the reference value by 23%, continuous monitoring is recommended" is output. Level 4, no fault detected: when At this time, no fault prompts are output, and the original signal sequences of each channel, the extracted feature vectors, and The values ​​are also stored in the diagnostic record database for subsequent offline retrospective analysis; For safety-critical faults, the aforementioned classification thresholds are uniformly lowered by 0.15. Safety-critical faults are defined by a preset system category field, which is stored in the fault code basic information table. Each fault code corresponds to a system category, including six types: braking system, steering system, airbag system, powertrain system, air conditioning system, and audio system. Fault codes belonging to any of the braking system, steering system, or airbag system categories are considered safety-critical faults; the rest are general faults. When a fault is classified as safety-critical, the first-level confirmed fault threshold is lowered from 0.85 to 0.70, the second-level suspected fault threshold is lowered from 0.70 to 0.55, and the third-level attention threshold is lowered from 0.50 to 0.35. For example, if the confidence level of an abnormal brake master cylinder pressure sensor signal is 0.62, it would only reach the third level of attention under general fault classification. However, because it is a safety-critical fault, the confidence level of 0.62 after the adjustment falls into the second-level suspected fault range, and a suspected fault alarm "Brake master cylinder pressure abnormal, it is recommended to check the brake fluid level and brake line sealing" is output, so that safety-critical faults can be responded to with a lower evidence threshold.

[0021] In this embodiment of the invention, real-vehicle diagnostic feedback is collected, and confirmed, misjudged, and missed judgment results are included in the training set. Incremental tuning of diagnostic parameters is periodically triggered to adapt to vehicle model differences and long-term changes in vehicle status. Specifically, this includes: After each diagnosis, the diagnostic equipment writes the optimal fault combination and classification results output by the diagnosis to the diagnostic log. The diagnostic log includes the diagnosis timestamp, vehicle identification number, list of output fault codes, and confidence level of each fault. And the classification level; after the vehicle repair is completed, the repair personnel confirm the diagnostic results one by one through the human-machine interface of the diagnostic equipment. For each output fault code, the repair personnel select one of the following three feedback labels: select "Confirmed" to indicate that the fault actually exists and has been eliminated by this repair; select "False positive" to indicate that the diagnostic equipment reported the fault but the fault does not exist after actual inspection; select "Missed" to indicate that the fault was not reported in this diagnosis but was found to exist during the repair process; for faults marked as "Missed" by the repair personnel, the diagnostic equipment automatically records the fault code and the severity of the fault confirmed during the repair, and at the same time, it backtracks to extract the feature vectors of each sensor channel at the time of the missed diagnosis. The value is used to store the missed detection record along with its feature data into the feedback database; The feedback database is triggered to perform incremental tuning of diagnostic parameters every 50 newly added feedback records or every 7 days. The targets of incremental tuning include the standard feature vectors of each fault template in the fault template library and the coefficients of the weight lower bound constraints in the linear programming model. and the effect coefficients of coupling constraints in the mixed integer programming model. and tolerance There are three types of parameters; the optimization process is as follows: The first type is to update the standard feature vector of the fault template library: For each fault type, extract all records marked as "confirmed" in the feedback database within the last 180 days, take the mean of each dimension of the feature vector of the same sensor channel associated with these records, and replace the corresponding dimension values ​​in the original standard feature vector with the mean value, so that the fault template gradually approaches the signal drift characteristics after long-term vehicle use. The second type involves updating the lower bound coefficient of the weights. The number of false positives and false negatives due to sensor signal quality within the past 180 days is statistically analyzed from the feedback database. False positives due to sensor signal quality refer to those recorded in the false positive logs for the corresponding sensor channels. Furthermore, the fault is still recorded with high confidence output. Missed detection due to sensor signal quality refers to the missed detection record corresponding to the sensor channel. However, records where the feature vector of this channel has a cosine similarity to the fault template greater than or equal to 0.6; if the number of such misclassifications exceeds twice the number of such missed classifications, it indicates that the current... Channels with excessively low weights, resulting in poor signal quality, may be missed due to insufficient weighting. Increase by 0.01; if the number of such missed detections exceeds twice the number of such false positives, it indicates that the current... Retaining excessive weight for channels with generally poor signal quality can lead to misjudgments. Lowered by 0.01; The adjustment range is limited to [0.02, 0.10]. The third category is updating the effect coefficient. With tolerance For each type of fault Each confirmation record extracts various observation indicators at the time of diagnosis. The actual deviation is calculated by taking the mean deviation of each observed indicator from all confirmed records of the fault over the past 180 days, and then dividing this mean by the range of the observed indicator to obtain the new deviation. Simultaneously calculate the standard deviation of these deviations, and use twice the standard deviation as the new standard deviation. This ensures that the tolerance range covers normal fluctuations. After the above incremental optimization is completed, the updated fault template and weight lower bound coefficient will be used. Effect coefficient and tolerance The package is a diagnostic parameter update package and pushed to all diagnostic devices of the same vehicle model via wireless network. The push adopts a gray-scale release strategy, initially pushing only to 10% of online devices. After the push, the confirmation rate of subsequent feedback from the 10% of devices is continuously monitored to see if it is lower than the preset 80% threshold. If the confirmation rate is not lower than 80% for 7 consecutive days, the full push is pushed to the remaining 90% of devices. If the confirmation rate is lower than 80%, the version is rolled back to the previous version and the optimization is marked as abnormal, awaiting manual review.

[0022] Example 2: This example introduces a device intelligent diagnostic system based on mixed integer optimization, which includes a signal confidence module, a weight fusion module, a coupling diagnostic module, a hierarchical output module, and a feedback optimization module; The signal confidence module is used to acquire the actual vehicle sensor signal stream, evaluate the signal quality channel by channel and extract fault features, and generate the initial confidence level of each sensor for candidate faults. The weighted fusion module is used to solve a linear programming problem for sensor fusion weights with signal quality assessment results as constraints, and obtain the weighted fusion confidence score. The coupled diagnostic module is used to input the weighted fusion confidence into the mixed integer programming model to solve for the optimal fault combination; The graded output module is used to grade and output the diagnostic results according to the confidence values ​​of each fault in the optimal fault combination. The feedback tuning module is used to collect real-vehicle diagnostic feedback, incorporate confirmed, misjudged, and missed results into the training set, and periodically trigger incremental tuning of diagnostic parameters.

[0023] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0024] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0025] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0026] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system and method described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0027] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways.

[0028] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A device intelligent diagnostic method based on mixed integer optimization, characterized in that: Includes the following steps: Acquire the actual vehicle sensor signal stream, evaluate the signal quality channel by channel and extract fault features, and generate the initial confidence level of each sensor for candidate faults; Using the signal quality assessment results as constraints, a linear programming problem is solved to obtain the weighted fusion confidence level by solving the sensor fusion weights. The weighted fusion confidence score is input into a mixed integer programming model to solve for the optimal fault combination. Based on the confidence values ​​of each fault in the optimal fault combination, the diagnostic results are output in a graded manner. Collect real-vehicle diagnostic feedback, and incorporate confirmed, misjudged, and missed results into the training set to periodically trigger incremental tuning of diagnostic parameters.

2. The intelligent device diagnostic method based on mixed integer optimization according to claim 1, characterized in that: The generation of initial confidence levels for each sensor regarding candidate faults includes: The signal quality index is calculated for each sensor channel, and the time-domain and frequency-domain features are extracted to form a feature vector. The feature vector is compared with the standard feature vector in the pre-set fault template library to obtain the cosine similarity. The product of the signal quality index and the cosine similarity is used as the initial confidence level.

3. The intelligent device diagnostic method based on mixed integer optimization according to claim 1, characterized in that: The linear programming method for solving the sensor fusion weights yields the weighted fusion confidence, including: The maximum value of the initial confidence of each sensor channel for each candidate fault is taken as the fault sensitivity. A linear programming model is constructed with the goal of maximizing the sum of the products of the fusion weight of each channel, the signal quality index, and the fault sensitivity. The model satisfies the constraints that the sum of the weights is 1, the weights are not less than the product of the preset proportional coefficient and the signal quality index and are not higher than 0.

5. The optimal fusion weight is obtained by solving the problem. For each candidate fault, the sum of the products of the optimal fusion weight of each channel, the signal quality index, and the initial confidence level is calculated as the weighted fusion confidence level.

4. The intelligent device diagnostic method based on mixed integer optimization according to claim 1, characterized in that: The step of inputting the weighted fusion confidence score into the mixed integer programming model to solve for the optimal fault combination includes: Obtain the weighted fusion confidence of candidate faults, the complexity coefficient of diagnostic rules, the list of mutual exclusion relationships and causal relationships between faults, as well as the effect coefficient, actual observed value and tolerance of each fault on the observed indicators. A mixed-integer programming model is constructed, with fault occurrence and rule activation as binary variables and severity as a continuous variable. The objective is to maximize the sum of the weighted fusion confidence and the product of the fault occurrence variable minus the cost of rule complexity and the cost of physical constraint violation. Mutual exclusion constraints, causal constraints, coupling constraints, severity correlation constraints and rule activation constraints are applied to solve for the optimal fault combination.

5. The intelligent device diagnostic method based on mixed integer optimization according to claim 4, characterized in that: The complexity coefficient is the sum of the number of sensor channels on which the diagnostic rule depends and the number of judgment steps, and the effect coefficient is the ratio of the deviation of the observed index to the range of the observed index when the fault occurs alone.

6. The intelligent device diagnostic method based on mixed integer optimization according to claim 1, characterized in that: The step of grading and outputting diagnostic results includes: The confidence level of each fault in the optimal fault combination is compared with a preset threshold to classify the faults into four levels: confirmed faults, suspected faults, faults requiring attention, and faults not detected. Safety-critical faults are defined by system categories, and the threshold values ​​for each level of safety-critical faults are uniformly lowered by 0.15 compared to general faults.

7. The intelligent device diagnostic method based on mixed integer optimization according to claim 6, characterized in that: The threshold for confirming a general fault is 0.85, the threshold range for suspected faults is [0.70, 0.85), the threshold range for faults requiring attention is [0.50, 0.70), and the threshold for faults not detected is less than 0.

50. The threshold for confirming a safety-critical fault is lowered to 0.70, the threshold for suspected faults is lowered to 0.55, and the threshold for faults requiring attention is lowered to 0.

35.

8. The intelligent device diagnostic method based on mixed integer optimization according to claim 1, characterized in that: The collected real-vehicle diagnostic feedback includes confirmed, misjudged, and missed diagnosis results in the training set, periodically triggering incremental tuning of diagnostic parameters, including: Maintenance personnel mark and confirm each fault code output by the diagnostic system, identifying misjudged or missed faults. For missed faults, they trace back the feature vector and signal quality indicators at that moment and store them in the feedback database. Incremental tuning is triggered when the feedback database adds 50 new records or when 7 days have passed since the last tuning.

9. The intelligent device diagnostic method based on mixed integer optimization according to claim 6, characterized in that: The incremental optimization includes updating the standard feature vector of the fault template with the mean of the feature vectors of the confirmed records in the most recent time period. Adjust the lower bound ratio coefficient of the weight based on the number of false positives and false negatives caused by signal quality; Update the effect coefficient and tolerance using the mean and standard deviation of the observed index deviations in the confirmation records.

10. A device intelligent diagnostic system based on mixed integer optimization, used to implement the device intelligent diagnostic method based on mixed integer optimization as described in any one of claims 1-9, characterized in that: It includes a signal confidence module, a weight fusion module, a coupling diagnosis module, a hierarchical output module, and a feedback optimization module; The signal confidence module is used to acquire the actual vehicle sensor signal stream, evaluate the signal quality channel by channel and extract fault features, and generate the initial confidence level of each sensor for candidate faults. The weighted fusion module is used to solve a linear programming problem for sensor fusion weights with signal quality assessment results as constraints, and obtain the weighted fusion confidence score. The coupled diagnostic module is used to input the weighted fusion confidence into the mixed integer programming model to solve for the optimal fault combination; The graded output module is used to grade and output the diagnostic results according to the confidence values ​​of each fault in the optimal fault combination. The feedback tuning module is used to collect real-vehicle diagnostic feedback, incorporate confirmed, misjudged, and missed results into the training set, and periodically trigger incremental tuning of diagnostic parameters.