Elevator fault diagnosis method and system

Abstract

The invention relates to the technical field of elevator fault diagnosis, and discloses an elevator fault diagnosis method and system.The elevator fault diagnosis method comprises the steps that elevator operation data are collected, operation stages are divided for the elevator operation process according to the elevator operation data, and the residual error of the elevator operation data is calculated in the operation stages; based on an elevator structure, a causal chain graph reflecting the relation between components is established through fault mode analysis, and a direct upstream node set of downstream nodes in the causal chain graph serves as a parent set of the downstream nodes; in the operation stage, performing block replacement processing for keeping time sequence characteristics on an upstream node residual sequence of each edge in the causal chain graph to generate a contrast residual sequence, and calculating the propagation intensity of the edge according to the contrast residual sequence; effective propagation paths are screened according to the propagation intensity of the edges, the effective propagation paths are sorted to obtain candidate nodes, residual errors of the candidate nodes are set to be zero through virtual pinch-off, forward propagation is carried out through a semi-physical model, and fault nodes are determined in the candidate nodes based on forward propagation results.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of elevator fault diagnosis, in particular to an elevator fault diagnosis method and system. BACKGROUND

[0002] With the continuous growth of the number and frequency of use of elevators, elevator safety and operational reliability are increasingly concerned. The traditional elevator fault diagnosis method mainly relies on the threshold judgment or simple logic trigger built in the controller, such as detecting motor overcurrent, door machine timeout or brake failure, etc. This kind of method can find obvious hardware failure, but it is insufficient for progressive deterioration, cross-coupling abnormality or early fault symptom identification under the linkage of multiple components. In addition, the threshold is fixed and lacks environmental adaptation, often resulting in false positives or false negatives, making it difficult to meet the increasingly complex operating scenarios.

[0003] In recent years, some research attempts to introduce data-driven or model-driven residual analysis methods, but most of them stay at the single variable or local subsystem level, lacking causal relationship modeling across multiple modules such as traction, braking, door machine, etc. The existing residual inspection methods are mostly based on single-point statistics or filtering estimation, which cannot effectively separate "real propagation" and "common noise", thus lacking reliability under multi-source interference. At the same time, common algorithms fail to combine with engineering priors (such as physical symbol relationship, time delay range, component action logic), resulting in a lack of explainability of the diagnosis results, making it difficult for maintenance personnel to directly accept and apply.

[0004] Therefore, there is an urgent need for an elevator fault diagnosis method that can face multiple data sources, combine semi-physical period modeling and residual causal chain inspection, and have a verifiable closed loop, which can not only realize early warning, but also output executable maintenance instructions. SUMMARY

[0005] In view of the above problems, the present application is proposed.

[0006] To solve the above technical problems, the present application provides the following technical scheme: an elevator fault diagnosis method, comprising: collecting elevator operation data, dividing the elevator operation process into operation stages according to the elevator operation data, and calculating the residual of the elevator operation data in the operation stage; Based on the structure of the elevator, a causal chain diagram reflecting the relationship between each component is established through fault mode analysis, and the direct upstream node set of the downstream node in the causal chain diagram is taken as the parent set of the downstream node; In the operation stage, the upstream node residual sequence of each edge in the causal chain diagram is subjected to block replacement processing to maintain the timing characteristics, to generate a contrast residual sequence, and the propagation strength of the edge is calculated according to the contrast residual sequence; The effective propagation path is screened according to the propagation strength of the edge, the candidate node is obtained by sorting the effective propagation path, the residual error of the candidate node is zeroed by using virtual pinch-off, and forward propagation is carried out through a semi-physical model, and the fault node is confirmed in the candidate node based on the forward propagation result.

[0007] As a preferred scheme of the elevator fault diagnosis method, the elevator operation data includes traction system data, environment and load data, door machine system data and brake system data. The traction system data includes current, DC bus voltage, car speed, car acceleration, motor output torque. The environment and load data includes car load estimation, car position floor information, environment temperature and humidity. The door machine system data includes current curve, action duration and door position sensor signal of the opening and closing process. The brake system data includes brake coil on-off current, brake release and brake action time.

[0008] As a preferred scheme of the elevator fault diagnosis method, the operation stage is divided into six stages: start-up acceleration stage, uniform speed operation stage, deceleration stage, leveling and braking stage, opening door stage and closing door stage, wherein the start-up acceleration stage, the uniform speed operation stage and the deceleration stage are traction-related operation stages.

[0009] As a preferred scheme of the elevator fault diagnosis method, the residual error of the elevator operation data in the operation stage includes, in the traction-related operation stage, the motor output torque expectation is estimated by simplifying the dynamic relationship, the motor output torque expectation is equal to the sum of the inertia demand torque, the viscous damping torque, the gravity equivalent torque and the friction torque, and the motor output torque expectation is mapped to the current expectation and the DC bus voltage expectation. In the leveling and braking stage, the brake coil on-off current expectation and the car speed expectation are constructed. In the opening and closing stage, based on the door machine system data, the action duration expectation and the current curve expectation of the opening and closing process are generated. The elevator operation data is subtracted from the corresponding expectation to obtain the residual error of the elevator operation data.

[0010] As a preferred embodiment of the elevator fault diagnosis method of the present invention, the causal chain diagram includes: using the residuals of the elevator operation data as nodes of the causal chain diagram, with one node corresponding to one residual; determining the causal relationship between the nodes based on the physical working principle of the elevator system; establishing directed edges from upstream causes to downstream results to complete the establishment of the causal chain diagram, wherein the node corresponding to the upstream cause is the upstream node, and the node corresponding to the downstream result is the downstream node; In the causal chain diagram, the upstream node directly connected to the downstream node through a directed edge is the direct upstream node of the downstream node, and the set of the direct upstream nodes of the downstream node is regarded as the parent set of the downstream node.

[0011] As a preferred embodiment of the elevator fault diagnosis method of the present invention, the block replacement process includes selecting a directed edge u→d in the causal chain graph during the operation phase, where u represents an upstream node, → represents the direction of the directed edge, and d represents a downstream node, extracting the residual of u and the residual of d during the operation phase, and simultaneously extracting the residual corresponding to the parent set of d; Eliminate the interpretable influence of d on the rest of the parent set, obtain the downstream new information related only to u, and construct the conditional innovation of the downstream node d, which is expressed as: ; in, represents the conditional innovation amount; represents the residual of d; Indicates downstream nodes In the parent set of , all residual sets except u; Indicates that in a given Under the conditions of Conventional estimates of Estimating upstream residuals The short-term correlation structure of The autocorrelation decay time is converted into the number of sample points in combination with the sampling frequency to obtain the block length, which is expressed as: ; in, represents the residual of u; Indicates the block length used for replacement; express The autocorrelation decay time of Indicates the sampling frequency; represents the ceiling operator; Perform cyclic block permutation to Divide into several continuous blocks, each block length L, randomly rearrange the order of blocks in units of blocks, and allow the end to end to form a ring arrangement. The time sequence within the block remains unchanged, and the permutation sequence is obtained. and Keeping unchanged, only replacing the upstream node u with the permutation sequence, the counterfactual control of the short-term structure of the upstream node being disturbed is constructed; The conditional innovation quantity is For the downstream response, the difference in the explanatory power of the conditional innovation quantity of the upstream node u and the permutation sequence is compared, and the conditional score of the upstream node u and the permutation sequence is calculated respectively using the conditional correlation score function, and the difference between the two times is taken as the propagation strength; B independent cyclic block permutations are performed to obtain the empirical distribution of the propagation strength, and the significance is calculated by permutation test, and B represents the permutation number with a positive integer value; In the running phase, the main lag position of the upstream node to the downstream node is estimated, and for the directed edge u→d, the propagation strength, significance and main lag position are output.

[0012] As a preferred scheme of the elevator fault diagnosis method, wherein: the sorting of the effective propagation path to obtain the candidate node includes, for each node in the effective propagation path, counting the edge set and the incoming edge set, the outgoing edge set is all downstream node edges pointed to by the node as upstream, and the incoming edge set is all edges pointed to it from other nodes as downstream; A root cause score value is defined for each node, and the root cause score value is composed of output contribution, input weakening and node residual energy, the output contribution is the sum of the propagation strength of all outgoing edges of the node, the input weakening is the sum of the propagation strength of all incoming edges of the node, and the node residual energy is the mean square of the residual of the node in the running phase, and the root cause score value is represented as the output contribution of the node minus the input weakening plus the node residual energy; The root cause scores of all nodes are calculated, sorted from high to low according to the root cause score value, and a root cause node sorting table is formed.

[0013] As a preferred scheme of the elevator fault diagnosis method, wherein: the virtual pinch-off includes selecting the first node in the root cause node sorting table as the target root cause, and taking out the outgoing edge set of the target root cause; Without changing other parent sets and running phases, the residual of the target root cause is artificially set to zero, and the magnitude of the downstream residual drop is predicted by forward propagation, and the corresponding prediction curve is generated, and for each downstream node, the predicted drop trajectory is compared with the current residual, and the predicted drop ratio is calculated.

[0014] An elevator fault diagnosis system using any of the methods described in the application, wherein: the acquisition module acquires elevator operation data, divides the elevator operation process into running phases according to the elevator operation data, and calculates the residual of the elevator operation data in the running phase; The analysis module, based on the elevator structure, establishes a causal chain diagram reflecting the relationship between various components through failure mode analysis, and takes the set of directly upstream nodes of the downstream node in the causal chain diagram as the parent set of the downstream node; The permutation module, during the running phase, performs block permutation processing on the residual sequence of the upstream node of each edge in the causal chain graph to maintain the timing characteristics, generates a control residual sequence, and calculates the propagation strength of the edge based on the control residual sequence; The positioning module screens effective propagation paths based on the propagation strength of the edges, sorts the effective propagation paths to obtain candidate nodes, uses virtual pinch-off to set the residuals of the candidate nodes to zero, and performs forward propagation through a semi-physical model. Based on the forward propagation results, the faulty node is confirmed among the candidate nodes.

[0015] Beneficial effects of the present invention: The method of the present invention can accurately identify abnormal relationships among subsystems such as traction, braking, and door machines under multi-source data through operation stage division, semi-physical expectation generation and residual construction, causal chain graph modeling, and block replacement propagation testing. Counterfactual controls are constructed while retaining short-term correlations, and the propagation intensity is quantified by conditional score differences to ensure that the causal chain is both statistically significant and in line with scientific and engineering logic. On this basis, root cause scoring is used to achieve node sorting and highlight source anomalies; and an observable closed loop is formed through virtual pinch-off and retest comparison, significantly improving the reliability, adaptability, and engineering operability of the diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 The present invention provides an overall flow chart of an elevator fault diagnosis method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0019] Example 1, reference Figure 1 , as an embodiment of the present invention, provides an elevator fault diagnosis method, comprising: S1: Collecting elevator operation data, dividing the elevator operation process into operation stages according to the elevator operation data, and calculating the residual of the elevator operation data in the operation stage.

[0020] The present application first establishes an analysis framework in stages, divides a complete elevator operation process into stages with clear physical meaning and repeatable identification, and manages multi-source data in each stage, so that subsequent analysis is compared under similar working conditions.

[0021] To support stage division and data slicing, multiple signals are collected through existing elevator control systems and conventional sensors, and a uniform and quality-controlled mechanism is formed through time stamping to form a neat multi-channel data stream. The system collects and synchronizes the following data in real time: motor three-phase current, DC bus voltage, car speed, car acceleration, motor output torque, brake coil on-off current, brake release and brake action time, current curve of door opening and closing process, action duration, door position sensor signal, car load estimate, car position floor information, ambient temperature and humidity. All channels are time-stamped, deburred, band-limited smoothed, missing data compensated and resampled aligned, and data quality identifiers are retained to constrain the sampling range of subsequent self-calibration.

[0022] According to the working logic and signal characteristics of the elevator mechanical and electrical system, the stage division is completed. The typical states of the elevator operation state include: starting from static to starting (the car speed continuously rises from zero, the motor three-phase current and the motor output torque are lifted synchronously), reaching and maintaining constant speed (the car speed is stable, the motor three-phase current fluctuation is limited), deceleration near the target floor (the car speed continuously decreases, the DC bus voltage shows a steady-state fluctuation pattern corresponding to regeneration or traction), leveling and braking (the car speed converges to the leveling range and the brake coil on-off current appears clear action), and opening and closing doors (the current curve of the opening and closing process presents a starting peak, a platform and an end alignment disturbance feature, and the door position sensor signal changes according to the expected timing).

[0023] The system uses a state machine to continuously distinguish the above stages on the time axis: the trend of speed and acceleration cooperates with the common change of motor three-phase current and motor output torque to determine starting, constant speed and deceleration; the steady voltage level and energy feedback characteristics of the DC bus voltage serve as the side evidence of the deceleration end and leveling; the action edge of the brake coil on-off current and the convergence of the car speed together determine the leveling and braking window; the opening and closing of the door is identified based on the current curve of the opening and closing process, the action duration and the logic of the door position sensor signal. For complex working conditions such as intermediate parking, re-leveling, and secondary start-stop, the state machine allows to set sub-states in the same major stage to mark the transition behavior, but all windows have clear stage labels and direction attributes (uplink / downlink).

[0024] When a phase boundary is identified, the system will cut a window of multi-channel data within the corresponding time range, with a small amount of transition margin on both ends to record the dynamics near the boundary completely. Each window carries the context information related to diagnosis, including direction, start and end floors (given by car position floor information), car load estimate, ambient temperature and humidity, whether energy feedback occurs, the specific actions of opening / closing doors and the timing of braking actions, etc. The above phase division and slicing process aligns the same working conditions under different dates, different seasons, and different passenger flows on a comparable basis, laying a stable data foundation for subsequent expectation generation and residual construction.

[0025] Further, within each operating phase, in combination with the conventional relationship between dynamics and electrical side, the expectation quantity is generated for the collected quantity related to the phase, and the residual is uniformly constructed with the deviation of "measured-expected". To avoid parameter rigidity, the parameters of all expectation quantities are based on the recent healthy window, and robust statistics with forgetting factor are used for self-calibration, maintaining the mobility under different equipment, different seasons and different load conditions.

[0026] Specifically, in the traction-related phase, the system generates the motor output torque expectation based on the simplified dynamics relationship, and then maps it to obtain the motor three-phase current expectation; the DC bus voltage expectation is given according to the nominal voltage target of the device, combined with the load state and temperature factors (used for early identification of voltage abnormality and feedback abnormality). The motor output torque expectation is represented as: ; Wherein, represents the motor output torque expectation; represents the equivalent moment of inertia; represents the angular acceleration, which is obtained by converting the car acceleration and transmission ratio or by differentiating the car speed value; represents the equivalent viscous damping coefficient; represents the angular velocity, which is obtained by converting the car speed and transmission ratio or estimated by the driver; represents the gravity equivalent term, which is estimated by the car load estimate, direction and floor displacement; represents the transmission friction equivalent term; represents time.

[0027] The motor output torque expectation is mapped to the current expectation and the DC bus voltage expectation, and the current expectation is represented as: ; Wherein, represents the current expectation; represents the torque constant; represents the equivalent bias of current measurement and control dead zone.

[0028] The DC bus voltage expectation is given according to the voltage stabilization target of the driver and the running state: the system takes the nominal voltage stabilization value of the device as the center, combines the slow drift caused by environmental temperature and humidity on the characteristics of the device, and the load change (indirectly reflected through the motor output torque expectation and the motor three-phase current expectation) to form a temperature-load corrected voltage stabilization expectation, and allows a controlled short-time surge or ripple bandwidth to occur in the deceleration scenario of regenerative braking; the expectation is used to compare with the DC bus voltage, so as to identify abnormal voltage stabilization deviation and feedback abnormalities.

[0029] In the leveling and braking phase, the system synchronously generates the brake coil on-off current expectation and the car speed expectation (converging in the last section). The former adopts a "rise-keep-decay" templated profile, and the parameters are obtained from the statistics of the healthy window and are fine-tuned according to the environmental temperature and humidity; the latter adopts a "uniform deceleration to the leveling speed range in the last section" convergence model, and the parameters are lightly corrected according to the car position floor information and the car load estimation value, so as to maintain causal consistency with the action edge of the brake coil on-off current.

[0030] In the opening and closing door phase, the system generates the action duration expectation and the current curve expectation of the opening and closing door process according to the timing and logic of the door position sensor signal. Among them, the action duration expectation adopts a "acceleration-constant speed-deceleration" three-segment kinematics template, and is slowly corrected according to the environmental temperature and humidity and the near-period use frequency (hot / cold); the current curve expectation of the opening and closing door process adopts a "start peak-plateau-end alignment disturbance" three-segment load profile, and the parameters are slowly updated according to the long-term guide rail cleanliness and door width characteristics, and are aligned with the key nodes of the door position sensor signal to constrain the timing of the end alignment.

[0031] It should be noted that in the flat layer, braking and door machine opening and closing stages, the "template profile" and "convergence model" adopted by the present application are both conventional engineering modeling methods in the field, and the core idea is to use simple and fixed function forms to fit and describe common execution component action processes. The "rise-maintain-decay" template profile refers to signals such as brake coil current, which has typical on-off characteristics, and its action process usually shows rapid current rise, maintenance in a certain amplitude range for a period of time, and then gradually decays to zero. The industry generally uses this three-section curve to represent the action process of electromagnetic brakes. The "final uniform deceleration to flat layer speed range" convergence model refers to the fact that the car will gradually reduce speed according to the law of uniform deceleration before reaching the station, until it approaches zero speed and enters the flat layer interval. This deceleration convergence model is the basis of elevator flat layer control and belongs to mature control law. The "acceleration-constant speed-deceleration" kinematics template is commonly used to describe the door machine opening and closing time curve, i.e. the door panel first accelerates to start, then runs at a substantially constant speed, and finally slows down to align near the end. The industry widely uses this three-section time model as a reference for opening and closing door performance evaluation. The "start peak-plateau-end alignment disturbance" current profile template corresponds to the load characteristics of the door machine on the electrical side, i.e. the current appears a peak value at start, maintains a plateau value during running, and has a small fluctuation at the end due to damping or alignment effect.

[0032] After generating the above-mentioned expected trajectory and event baseline, the system constructs a unified residual representation within the current window. Specifically, in each operating phase, all residuals are defined as the difference between the elevator operating data and the corresponding expected value, represented as: Motor three-phase current residual = motor three-phase current - motor three-phase current expected; DC bus voltage residual = DC bus voltage - DC bus voltage expected; Car speed residual = car speed - car speed expected (flat layer section); Car acceleration residual = car acceleration - car acceleration expected; Motor output torque residual = motor output torque - motor output torque expected; Brake coil on-off current residual = brake coil on-off current - brake coil on-off current expected; Action duration residual = action duration - action duration expected; Current curve residual of opening and closing process = current curve of opening and closing process - current curve of opening and closing process expected.

[0033] To achieve comparability between different dimensions and different phases, the system normalizes the residuals. Normalization does not change the mapping relationship between expectation and residual, but puts residuals of different dimensions on a unified scale, facilitating threshold setting, statistical testing and cross-dimension comparison in subsequent steps. Both residuals and normalized residuals are stored together with the context information of the operating phase (direction, car position floor information, car load estimate, ambient temperature and humidity, data quality identifier) for subsequent diagnostic reasoning and model self-updating.

[0034] S2: Based on the elevator structure, a causal chain graph reflecting the relationship between each component is established through failure mode analysis, and the direct upstream node set of the downstream node in the causal chain graph is taken as the parent set of the downstream node.

[0035] Further, after completing the phasing and residual construction, the causal chain graph covering the traction system, braking system, door machine system and electrical energy supply link is established based on the physical mechanism of the elevator mechanical and electrical system, control logic and maintenance experience. The nodes of the causal chain graph are represented by residuals, i.e. the motor three-phase current residual, DC bus voltage residual, car speed residual, car acceleration residual, motor output torque residual, brake coil on-off current residual, action duration residual, current curve residual of opening and closing door process, etc. To improve the explainability and stability of causal relationship, the establishment of the graph follows the principle of "physical cause first, signal consequence second": the directionality is determined in priority according to the mechanical, electrical and control links between components, supplemented by known associations in the maintenance manual, FMEA failure mode analysis conclusions and statistical confirmation of historical health samples. For different phases (start, constant speed, deceleration, leveling and braking, opening door, closing door) and different directions (up, down), the graph structure allows slight differences, and the system maintains corresponding sub-graphs to separately depict the causal paths specific to each phase.

[0036] In the traction-related phase, the motor output torque residual is regarded as the upstream driving quantity of the traction link. If the motor output torque residual deviates, it will first be reflected in the motor three-phase current residual, and further affect the fluctuation of the DC bus voltage residual and the DC side power through energy transmission; at the same time, there is a causal relationship between the motor output torque residual and the car speed residual, car acceleration residual, especially in the start and deceleration phases. Based on this, the system sets a directed edge from the motor output torque residual to the motor three-phase current residual, a directed edge from the motor three-phase current residual to the DC bus voltage residual in the traction sub-graph, and according to the characteristics of the phase, sets directed edges from the motor output torque residual to the car speed residual, from the car speed residual to the DC bus voltage residual, etc.

[0037] For the braking phase, the braking coil on-off current residual directly affects the convergence behavior of the last segment of the car speed residual, and has an observable impact on the DC bus voltage residual during regenerative braking. Therefore, a directed edge is set from the braking coil on-off current residual to the car speed residual and the DC bus voltage residual in the braking subgraph.

[0038] For the door machine system, the current curve residual of the opening and closing process reflects the deviation of the door machine load and driving state, which can cause the action duration residual to systematically lengthen or shorten, and eventually present abnormalities such as leading, lagging, or jitter in the door position sensor signal residual. Therefore, a directed edge is established in the door machine subgraph in the order of “current curve residual of the opening and closing process → action duration residual → door position sensor signal residual”.

[0039] The direction and existence of the above directed edges are based on the device structure and control flow, and the rationality of the directed edges can be statistically verified in combination with historical health samples. To avoid unnecessary complexity, the graph structure is designed in a top-down hierarchical manner, and loops are minimized; if necessary, the intrinsic coupling relationship is preserved, and the influence is weakened by staging and timing constraints.

[0040] In the causal chain graph, through a directed edge, the upstream node directly connected to the downstream node is the direct upstream node of the downstream node. The direct upstream node set of each node is the parent set of the node. The parent set only contains the residuals that have a direct causal impact on the node, avoiding the introduction of remote variables that span multiple intermediate links, to ensure the explicitness of the explanation and the controllability of the calculation. For example, the parent set of the motor three-phase current residual usually only contains the motor output torque residual; the parent set of the DC bus voltage residual contains the motor three-phase current residual and the car speed residual; the parent set of the action duration residual contains the current curve residual of the opening and closing process; the parent set of the door position sensor signal residual contains the action duration residual; the parent set of the car speed residual contains the motor output torque residual during the starting and deceleration phases, and the braking coil on-off current residual during the leveling and braking phases. For the ambient temperature and humidity and the car load estimate, the system regards them as slow variables that have been corrected on the expectation side and does not directly include them in the parent set of the residual node.

[0041] To enhance the engineering constraints of the parent-child relationship, the system simultaneously registers the expected influence sign and the allowed time delay range on each directed edge. The influence sign is used to express the typical directional relationship of the upstream residual to the downstream residual, such as the increase in the motor output torque residual usually leading to the increase in the motor three-phase current residual; the allowed time delay range reflects the time window required for mechanical, electrical or control link propagation, such as the influence of the brake coil on-off current residual to the car speed residual usually occurring within a limited time delay of milliseconds to seconds. These information comes from the device structure, control cycle and statistical estimation of historical samples, which will be used as consistency gating conditions in subsequent steps. For nodes with multiple candidate upstreams, the system preferentially retains the upstream that can be explicitly explained by physical mechanisms, and if there is a conflict of equivalent explanations, the priority is set according to the stage characteristics and maintenance experience, and if necessary, the directed edges with low priority are disabled to ensure the sparsity and stability of the parent set.

[0042] Further, after the parent set is determined, the system constructs a conditional innovation quantity for each node within each running stage. The so-called conditional innovation quantity refers to the remaining part of the node residual that cannot be explained after considering the influences that can be explained by its parent set. In engineering implementation, the system first uses the parent set residual to perform the regular conditional elimination processing on the current node residual within the same window, such as eliminating the trend, amplitude and hysteresis components driven by the parent set, and then the remaining unexplained components are taken as the conditional innovation quantity of the node in the current window.

[0043] It should be noted that the conditional innovation quantity represents the new abnormal information of the node relative to its direct upstream, reducing the redundancy brought by upstream propagation; on the other hand, it provides a clear contrast for subsequent propagation inspection, so that the system can judge whether the disturbance of a certain upstream residual has left a repeatable influence trace on the downstream node. For windows with poor data quality (such as obvious missing, saturation or time misplacement), the system will limit the window from participating in elimination and contrast before constructing the conditional innovation quantity according to the quality identifier, so as not to mistake noise as unexplained components.

[0044] S3: In the running stage, perform block replacement processing on the upstream node residual sequence of each edge in the causal chain diagram to generate a contrast residual sequence, and calculate the propagation strength of the edge according to the contrast residual sequence.

[0045] Further, after the causal chain graph is established and the parent set and conditional innovation of each node are obtained, the propagability of any directed edge is checked in the running stage. Specifically, if an upstream residual is truly driving a downstream residual, when we only change the time structure of the upstream and keep other quantities in the parent set and downstream quantities unchanged, the downstream should show significant differences in its conditional innovation; otherwise, if the differences disappear, it is more likely to be co-driving or accidental coherence. The present application thus constructs the original- counterfactual contrast, and the difference in explanatory power of the two under the same parent set conditions is defined as the propagation strength of the edge.

[0046] In the running stage, a directed edge u→d in the causal chain graph is selected, u represents the upstream node, → represents the direction of the directed edge, and d represents the downstream node. The residual of u and the residual of d are taken from the running stage, and the residuals corresponding to the parent set of d are also taken out to eliminate the explainable influence of d on the remaining parent set, obtaining new information of the downstream related only to u. The conditional innovation of the downstream node d is constructed and represented as: ; wherein, represents the conditional innovation; represents the residual of d; represents the parent set of the downstream node except u; represents the conventional estimated value of under the condition of .

[0047] To make the "counterfactual" both disrupt the short-term time structure of the upstream and not destroy its amplitude distribution and intra-block correlation, the present application adopts a cyclic block permutation matching the autocorrelation time of the upstream residual. Specifically, the autocorrelation decay time of the upstream residual is first measured in the current window, and the block length is obtained by converting it with the sampling frequency. The permutation rearranges the blocks as the minimum unit, and the intra-block time sequence remains completely unchanged. The block length L is given by: ; wherein, represents the residual of u; represents the block length for permutation; represents the autocorrelation decay time of ; represents the sampling frequency; represents the upward rounding operator. The block length determined adaptively according to the correlation time can preserve the short-term spectral shape and autocorrelation of the upstream, thereby constructing a more realistic counterfactual contrast of what the upstream driving will be like when it is disturbed. This is particularly crucial for periodic disturbances, cogging effects, and door machine rhythms in elevator sites.

[0048] After determining the block length, only the upstream residual of that directed edge is permuted, while all other parent set residuals and downstream residuals remain unchanged. This precisely limits the permutation to a single directed edge in the causal chain. For example, in the traction link from estimated torque residual to motor three-phase current residual, only the estimated torque residual is permuted, leaving the motor three-phase current residual and other quantities in its parent set unchanged. In the braking link from brake coil on-off current residual to car speed residual, only the brake coil on-off current residual is permuted, leaving the car speed residual and its other parent sets unchanged. This directed edge localization avoids the confusion caused by traditional integer variable permutations that disrupt multiple paths, ensuring that propagation checks strictly correspond to the validity of this directed edge.

[0049] Specifically, perform cyclic block permutation to Divide into several continuous blocks, each block length L, randomly rearrange the order of blocks in units of blocks, and allow the end to end to form a ring arrangement. The time sequence within the block remains unchanged, and the permutation sequence is obtained. and Keep it unchanged, only replace the upstream node u with a permutation sequence, and construct a counterfactual control in which the short-term structure of the upstream node is disrupted.

[0050] To quantify the original-counterfactual difference, the conditional innovation For downstream response, the difference in explanatory power of the upstream node u and the permutation sequence on the conditional innovation quantity is compared, and the conditional correlation scoring function is used to calculate the two conditional scores of the upstream node u and the permutation sequence respectively, and the difference between the two conditional scores is taken as the transmission intensity.

[0051] Conditional correlation scoring functions can be implemented using standard methods such as conditional distance correlation, conditional mutual information, or partial-HSIC. Causal propagation is distinguished from common drivers / synchronous noise using the conditional score difference. Because the parent set conditions are fixed and the downstream explained components have been eliminated, the propagation strength will only be significantly positive if the upstream temporal structure truly leaves a reproducible impact downstream; otherwise, the difference will be close to zero.

[0052] Taking into account the influence of on-site noise and occasional impacts, we introduce multiple independent permutations to obtain empirical values ​​and give a significance measure based on them to avoid the randomness of a single comparison. By performing B independent cyclic block permutations on the upstream residuals, we obtain a set of propagation strength samples. The empirical value is calculated according to the conventional permutation test method. When the empirical value is small enough and remains stable within the adjacent window, it is confirmed that the directed edge has statistically significant propagation, which is expressed as: ; in, Indicates experience value; represents the number of permutations, which is a positive integer; b represents the bth permutation; denotes the propagation strength obtained by the b-th permutation; denotes the initial propagation strength; denotes the indicator function, which takes the value 1 when the condition in the brackets is true.

[0053] To consistently synchronize the symbol - time-lag consistency gate with the subsequent, the dominant time-lag position from upstream to downstream is estimated synchronously within the running phase (e.g. brake links usually take effect in the range of milliseconds to seconds, while traction links take effect in the order of control period). The dominant time-lag position is denoted as: ; wherein, denotes the dominant time-lag position; denotes the independent variable corresponding to the maximum value; denotes the candidate time-lag position; denotes the time-lag window allowed at the device level; denotes the correlation coefficient; denotes the upstream node residual at time The dominant time-lag will enter the gate along with the device-prior allowed time-lag window and the expected influence symbol of the directed edge, to filter out statistically significant but engineering unfeasible pseudo-propagations.

[0054] In engineering implementation, the present application gives application examples for several typical directed edges: In the directed edge of estimating torque residual → motor three-phase current residual, the propagation strength is significant and the dominant time-lag falls within the control period, usually indicating insufficient compensation on the drive side or deviation in friction and load estimation; In the directed edge of brake coil on-off current residual → car speed residual, if the propagation strength is significant and the dominant time-lag appears immediately along the directed edge, it indicates that the brake hysteresis or brake gap is abnormal; In the directed edge of current curve residual during door opening and closing process → action duration residual, if the propagation strength is significant and intensifies in the terminal alignment stage, it usually corresponds to poor guide rail lubrication or door slider wear. These conclusions are derived from the comparison and quantification of block permutation with conditional score difference, rather than a single threshold criterion, and thus have better interpretability and retest consistency.

[0055] To adapt to different devices and environments, the present application sets up robust protection for abnormal or degraded scenarios: when the upstream residual is approximately white noise and no reliable correlation time can be obtained, the block length is limited to the minimum available range and the confidence weight of the propagation conclusion is reduced; when the quality identifier of the stage window is unqualified (e.g. there is obvious missing sampling or clock misalignment), the propagation result of this window does not participate in aggregation; when the upstream or downstream is in saturation or clipping state, the permutation is suspended and resumed in the next healthy window. The above protection strategies ensure that statistical inference is not "biased" by low-quality data.

[0056] In summary, the present application forms a propagation verification mechanism that can be run in real-time at the edge and can be reviewed by engineers through three elements: time-adaptive block length, cyclic block permutation, and original-anti-factual conditional score difference. The output of the propagation strength, empirical value, and dominant time lag will directly enter the subsequent symbol-time lag consistency gating and root cause score aggregation, providing reliable and traceable evidence chain for early identification and root cause localization of elevator failures without changing the existing acquisition and control architecture.

[0057] S4: Screening effective propagation paths according to the propagation strength of the edges, sorting the effective propagation paths to obtain candidate nodes, setting the candidate node residuals to zero by virtual clipping and performing forward propagation through a semi-physical model, and confirming the fault nodes in the candidate nodes based on the forward propagation results.

[0058] Further, after completing the residual propagation verification, significance determination, and engineering prior gating, the present application obtains a purified causal chain graph. In this graph, each node corresponds to the residual of the aforementioned acquisition data, such as motor three-phase current residual, DC bus voltage residual, car speed residual, car acceleration residual, estimated torque residual, brake coil on-off current residual, action duration residual, current curve residual of door opening and closing process, and door position sensor signal residual, etc.; each directed edge represents a statistically significant, physically reasonable propagation relationship between upstream and downstream residuals.

[0059] In constructing the purified causal chain graph, a strict three-fold judgment method is adopted for the retention of directed edges: first, the propagation strength must be greater than zero and reach a preset threshold within the phase window, ensuring that the propagation is not random noise; second, this propagation strength needs to remain significant under multiple cyclic block permutations, i.e., the empirical value is less than the significance level (such as 0.05), otherwise the directed edge is considered not to exist; finally, the dominant time lag and sign of the propagation must meet the engineering prior, for example, the estimated torque residual → motor three-phase current residual should be positively correlated and the time lag should not exceed the drive control period. Only directed edges that meet all three conditions will be retained for the subsequent scoring process, and the remaining directed edges will be excluded from the graph. This processing ensures that the retained propagation relationships are statistically significant and comply with device physical laws, thereby avoiding false positives caused by accidental correlation or abnormal noise.

[0060] After obtaining the purified causal chain diagram, the root cause degree of each node is quantified. Specifically, the system first counts the sum of the propagation strengths of all outgoing edges of the node as the contribution of the node to the external diffusion of anomalies; meanwhile, the sum of the propagation strengths of all incoming edges is counted as the degree of dependence of the node on the upstream. The incoming edge refers to the directed edge pointing to the node, and the outgoing edge refers to the directed edge pointing from the node to other nodes. If a certain node has a significant propagation impact on the downstream, and itself is less affected by the upstream, the root cause possibility is higher. In order to avoid one-sided judgment by the propagation relationship alone, the present application further introduces the residual energy of the node itself as a supplementary index, that is, the energy or mean square value of the residual of the node is calculated within the stage window, to measure the magnitude of the node deviating from the expectation. In this way, the root cause score of each node is composed of three parts: external contribution, internal dependence, and residual energy.

[0061] In actual operation, the root cause score is calculated for all nodes according to the above rules, and is sorted according to the score. The node with the highest score often shows strong residual energy and can propagate anomalies to multiple downstream nodes, and its incoming edge is weak, indicating that the node is more likely to be the root cause. For example, if the estimated torque residual continuously produces significant propagation to the motor three-phase current residual and the DC bus voltage residual in multiple windows, and the residual energy of the node itself is also large, while the incoming edge has limited influence, the system will determine it as a potential root cause of the traction system; similarly, if the brake coil on-off current residual has a stable influence on the car speed residual and the DC bus voltage residual, the brake system can be marked as a suspicious source. Finally, the present application outputs a root cause node ranking table, which lists the root cause score, ranking and key propagation path of each node, for maintenance personnel to quickly locate the fault starting point.

[0062] Further, after completing the root cause node scoring and sorting, based on the size of the node score, the range of the propagation path and the strength of the residual energy, a fault alarm is further generated and processed. Specifically, the system marks the nodes with high root cause score and stable in multiple windows as high-risk sources, combines the breadth of the propagation range and the number of affected downstream nodes, and divides the anomaly event into different levels, for example: level I alarm corresponds to a root cause node with significant safety hazards, extensive propagation and significant residual energy; level II alarm corresponds to a node with functional degradation, limited propagation or medium residual energy; level III prompt is used for early deviation or slight anomaly nodes.

[0063] After the ranking, the system automatically generates maintenance guidelines for maintenance personnel by combining the physical meaning in the causal chain diagram and the engineering knowledge base. Specifically, if the estimated torque residual in the traction chain link is marked as the root cause, the system will prompt to check the motor excitation, driver parameters or load estimation link; if the brake coil on-off current residual appears high-level alarm, it prompts to check the brake coil, brake clearance or control relay; if the current curve residual of the door opening and closing process is continuously abnormal, it prompts to check the door machine motor, guide rail lubrication state or door lock mechanism. The maintenance guide content includes not only the components that need to be paid attention to, but also the suggested detection order and alternative quick troubleshooting methods, so that the alarm information can directly guide operation and maintenance.

[0064] Furthermore, after the root cause positioning is completed, the system does not directly use statistical significance as the final basis, but gives maintenance personnel a quantitative rollback commitment that can be verified by retesting through virtual clipping.

[0065] Virtual clipping refers to artificially setting the residual of the root cause determined to the health baseline in the online data space without changing the actual equipment, and deriving its natural influence on downstream residuals accordingly. Specifically, the first node in the root cause node ranking table is selected as the target root cause , the out-edge set of the target root cause is taken out, the residual of the target root cause is artificially set to zero without changing other parent sets and operating stages, and the magnitude of the downstream residual rollback is predicted by forward propagation, and the corresponding prediction curve is generated, represented as: ; Wherein, represents the residual prediction value of downstream node d after the root cause node is virtually clipped; represents the measured residual value of downstream node d; represents the target root cause The interpretable component caused to the downstream node d by the directed edge.

[0066] To make the commitment comparable, the system records the context constraints related to the window, including floor pair (given by the car position floor information), direction, car load estimate, ambient temperature and humidity, and whether energy feedback occurs, at the same time when generating the prediction rollback list. Maintenance personnel arrange retests under similar conditions according to this, to ensure that the windows before and after maintenance are comparable. After the retest is completed, the system takes the residual energy comparison before and after maintenance as the objective acceptance standard: when the predicted rollback is verified on the key downstream residual, and occurs in the key period consistent with the prediction (such as the end of deceleration, the end of door alignment, etc.), the system can determine that the root cause is established; if the rollback is insufficient or the period is not consistent, the system automatically reverts to the alternative root cause, or prompts to perform group action verification. To avoid misjudgment caused by noise or single occurrence, the system performs consistency check on multiple adjacent windows: only when the rollback amplitude and period are stable and repeated in adjacent windows, the maintenance is considered to be passed.

[0067] Through the above closed-loop mechanism, the application closely connects statistical inference with engineering practice: virtual pinch turns the propagation conclusion of upstream-downstream into measurable, comparable, and verifiable rollback commitment; retest comparison provides objective acceptance standard for alarm and maintenance.

[0068] In an exemplary embodiment, an elevator fault diagnosis system is also provided, including a collection module that collects elevator operation data, divides an elevator operation process into operation stages according to the elevator operation data, and calculates residuals of the elevator operation data in the operation stages.

[0069] An analysis module that, based on the elevator structure, establishes a cause-effect chain diagram reflecting the relationship between components by fault mode analysis, and takes the direct upstream node set of a downstream node in the cause-effect chain diagram as the parent set of the downstream node.

[0070] A replacement module that, in the operation stages, performs block replacement processing that maintains the timing characteristics on the upstream node residual sequence of each edge in the cause-effect chain diagram, generates a comparison residual sequence, and calculates the propagation strength of the edge according to the comparison residual sequence.

[0071] A positioning module that filters effective propagation paths according to the propagation strength of the edge, sorts the effective propagation paths to obtain candidate nodes, sets the residuals of the candidate nodes to zero by virtual pinch and performs forward propagation through a semi-physical model, and confirms the fault node in the candidate nodes based on the forward propagation result.

[0072] If the above functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or parts of the present application that make contributions to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0073] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a list of executable instructions for implementing logic functions, which can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus or device, such as a computer-based system, a system including a processor or other system that can fetch instructions from an instruction execution system, apparatus or device and execute the instructions, or in conjunction with these instruction execution systems, apparatus or devices. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transport programs for use by or in connection with an instruction execution system, apparatus or device, or in conjunction with these instruction execution systems, apparatus or devices.

[0074] More specific examples (non-exhaustive list) of the computer-readable medium include the following: an electrical connection having one or more wires (electrical devices), a portable computer diskette (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CD ROM). In addition, the computer readable medium can even be paper or other suitable medium on which the program can be printed, because the program can be obtained electronically, for example, by optical scanning of the paper or other medium, followed by editing, interpreting or otherwise processing, if necessary, in other suitable ways, to be electronically obtained and then stored in the computer memory.

[0075] It should be understood that portions of the present application can be implemented in hardware, software, firmware, or combinations thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, implementation can be with any or a combination of the following technologies, which are all well known in the art: a discrete logic circuit having logic gates for implementing logic functions upon an application of data signals, an application specific integrated circuit having appropriate combinational logic gates, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0076] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and all of them should be covered in the scope of the claims of the present application.

Claims

1. A method for diagnosing elevator faults, characterized in that: include: Collect elevator operation data, divide the elevator operation process into operation stages according to the elevator operation data, and calculate the residual of the elevator operation data within the operation stage; Based on the elevator structure, a causal chain diagram reflecting the relationship between various components is established through failure mode analysis. The set of directly upstream nodes of the downstream node in the causal chain diagram is taken as the parent set of the downstream node. During the running phase, the residual sequence of the upstream node of each edge in the causal chain graph is subjected to block permutation processing to maintain the timing characteristics, generating a control residual sequence, and calculating the propagation strength of the edge based on the control residual sequence; Valid propagation paths are screened according to the propagation strength of the edges, and candidate nodes are obtained by sorting the valid propagation paths. The residuals of the candidate nodes are set to zero using virtual pinch-off and forward propagation is performed through a semi-physical model. The faulty node is confirmed among the candidate nodes based on the forward propagation results.

2. An elevator fault diagnosis method according to claim 1, characterized in that: The elevator operation data includes traction system data, environment and load data, door machine system data and brake system data; The traction system data includes current, DC bus voltage, car speed, car acceleration, and motor output torque; The environment and load data include the estimated value of the car load, the floor information of the car position, and the ambient temperature and humidity; The door machine system data includes the current curve of the door opening and closing process, the action duration and the door position sensor signal; The braking system data includes the on-off current of the brake coil, the brake release and the brake action time.

3. An elevator fault diagnosis method according to claim 2, characterized in that: The operation stages are divided into six stages: starting acceleration stage, uniform speed operation stage, deceleration stage, leveling and braking stage, door opening stage and door closing stage, among which the starting acceleration stage, uniform speed operation stage and deceleration stage are traction-related operation stages.

4. An elevator fault diagnosis method according to claim 3, characterized in that: Calculating the residual of the elevator operation data during the operation phase includes, during the traction-related operation phase, estimating the expected motor output torque by simplifying the dynamic relationship, where the expected motor output torque is equal to the sum of the inertia demand torque, the viscous damping torque, the gravity equivalent torque, and the friction torque, and mapping the expected motor output torque to the expected current and the expected DC bus voltage; During the leveling and braking phases, the expected on-off current of the brake coil and the expected car speed are constructed; During the door opening and closing phases, based on the door machine system data, the expected action duration and the expected current curve of the door opening and closing process are generated; Subtract the corresponding expectation from the elevator operation data to obtain the residual of the elevator operation data.

5. An elevator fault diagnosis method according to claim 4, characterized in that: The establishment of a causal chain diagram reflecting the relationship between the various components through failure mode analysis includes: using the residuals of the elevator operation data as nodes of the causal chain diagram, with one node corresponding to one residual; determining the causal relationship between the nodes based on the physical working principle of the elevator system; establishing directed edges from the upstream cause to the downstream result, and completing the establishment of the causal chain diagram, wherein the node corresponding to the upstream cause is the upstream node, and the node corresponding to the downstream result is the downstream node; In the causal chain diagram, the upstream node directly connected to the downstream node through a directed edge is the direct upstream node of the downstream node, and the set of the direct upstream nodes of the downstream node is regarded as the parent set of the downstream node.

6. An elevator fault diagnosis method according to claim 5, characterized in that: The block replacement process includes selecting a directed edge u→d in the causal chain graph during the running phase, where u represents an upstream node, → represents the direction of the directed edge, and d represents a downstream node, extracting the residual of u and the residual of d during the running phase, and simultaneously extracting the residual corresponding to the parent set of d; Eliminate the interpretable influence of d on the rest of the parent set, obtain the downstream new information related only to u, and construct the conditional innovation of the downstream node d, which is expressed as: ； in, represents the conditional innovation amount; represents the residual of d; Indicates downstream nodes In the parent set of , all residual sets except u; Indicates that in a given Under the conditions of Conventional estimates of Estimating upstream residuals The short-term correlation structure of The autocorrelation decay time is converted into the number of sample points in combination with the sampling frequency to obtain the block length, which is expressed as: ； in, represents the residual of u; Indicates the block length used for replacement; express The autocorrelation decay time of Indicates the sampling frequency; represents the ceiling operator; Perform cyclic block permutation to Divide into several continuous blocks, each block length L, randomly rearrange the order of blocks in units of blocks, and allow the end to end to form a ring arrangement. The time sequence within the block remains unchanged, and the permutation sequence is obtained. and Keep it unchanged, only replace the upstream node u with a permutation sequence, and construct a counterfactual control in which the short-term structure of the upstream node is disrupted; Conditional innovation For downstream response, the difference in explanatory power of upstream node u and permutation sequence on conditional innovation is compared, and the conditional correlation score function is used to calculate the two conditional scores of upstream node u and permutation sequence respectively, and the difference between the two conditional scores is taken as the transmission intensity; Perform B independent cyclic block permutations to obtain the empirical distribution of propagation strength and calculate the significance through permutation test, where B represents the number of permutations, which is a positive integer. During the running phase, the main lag position of the upstream node to the downstream node is estimated, and for the directed edge u→d, the propagation strength, significance and main lag position are output.

7. An elevator fault diagnosis method according to claim 6, characterized in that: Sorting the valid propagation paths to obtain candidate nodes includes, for each node in the valid propagation path, counting an outgoing edge set and an incoming edge set, wherein the outgoing edge set is all downstream node edges pointed to by the node when the node is an upstream node, and the incoming edge set is all edges pointed to the node from other nodes when the node is a downstream node; A root cause score is defined for each node. The root cause score consists of output contribution, input attenuation, and the node's own residual energy. Output contribution is the sum of the propagation strengths of all outgoing edges of the node. Input attenuation is the sum of the propagation strengths of all incoming edges of the node. The node's own residual energy is the mean square of the residual of the node calculated during the runtime. The root cause score is expressed as the node's output contribution minus input attenuation plus the node's own residual energy. Calculate the root cause scores for all nodes and sort them from high to low according to the root cause scores to form a root cause node sorting table.

8. An elevator fault diagnosis method according to claim 7, characterized in that: The virtual pinch-off includes selecting the first-ranked node in the root cause node ranking table as the target root cause, and extracting the outgoing edge set of the target root cause; Without changing other parent sets and operation stages, the residual of the target root cause is artificially set to zero, and the magnitude of the downstream residual drop is predicted by forward propagation, and the corresponding prediction curve is generated. For each downstream node, the predicted drop trajectory is compared with the current residual, and the predicted drop ratio is calculated.

9. An elevator fault diagnosis system, applied to an elevator fault diagnosis method according to any one of claims 1 to 8, characterized in that: include, The acquisition module collects elevator operation data, divides the elevator operation process into operation stages according to the elevator operation data, and calculates the residual of the elevator operation data within the operation stage; The analysis module, based on the elevator structure, establishes a causal chain diagram reflecting the relationship between various components through failure mode analysis, and takes the set of directly upstream nodes of the downstream node in the causal chain diagram as the parent set of the downstream node; The permutation module, during the running phase, performs block permutation processing on the residual sequence of the upstream node of each edge in the causal chain graph to maintain the timing characteristics, generates a control residual sequence, and calculates the propagation strength of the edge based on the control residual sequence; The positioning module screens effective propagation paths according to the propagation strength of the edges, sorts the effective propagation paths to obtain candidate nodes, uses virtual disconnection to set the residuals of the candidate nodes to zero, and performs forward propagation through the semi-physical model. Based on the forward propagation results, the faulty node is confirmed among the candidate nodes.

Images