A port crane mechanism degradation modeling and residual life estimation method and system

Abstract

The application provides a port crane mechanism degradation modeling and residual life estimation method and system, and relates to the technical field of port loading and unloading machinery health monitoring and predictive maintenance. The method takes the steady-state output section and the working condition Tag as input, establishes the same and similar working condition judgment rules to construct the same working condition comparable sequence; based on the health reference subset, the working condition is corrected and the non-degradation abnormal samples are removed by using the port machine operation event anchor point; multi-working condition partition modeling is performed in the complex mode; the sequence is subjected to joint change point detection and the degradation model is bound according to the change point type to generate the future health evolution path; according to the continuous overrun criterion and the same or similar working condition window arrival rate, the residual life is converted, and the inspection adjustment, load limiting or shutdown strategy signal is output. The application can distinguish the working condition change, non-degradation transient and real degradation, and improve the engineering reliability of residual life estimation.

Description

Technical Field

[0001] This invention relates to port loading and unloading machinery health monitoring and predictive maintenance technology, specifically to a method and system for port crane mechanism degradation modeling and remaining life estimation. Background Technology

[0002] Port cranes (such as quay container cranes and rail-mounted gantry cranes) are characterized by frequent starts and stops, large load switching, multi-mechanism linkage, and frequent changes in operating speed and position range during operation. Taking the hoisting mechanism of a quay container crane as an example, a complete loading and unloading cycle involves alternations between no-load, heavy-load, different speed gears, different lifting height ranges, and braking and linkage states. This results in long-term strong non-steady-state and mixed multi-condition monitoring data for vibration, temperature, current, and other conditions.

[0003] Currently, when monitoring the degradation and analyzing the lifespan of crane mechanisms, continuous monitoring data is typically arranged along natural time or segmented into equal-length segments. Trend extraction and lifespan prediction are then performed directly based on the constructed time series. However, because the load, speed, position, operating stage, and linkage state corresponding to different data segments are not consistent, the samples in the constructed feature series lack comparability of operating conditions, and degradation trends are easily masked by these differences. For example, the vibration intensity under heavy-load, uniform-speed upward operation is naturally higher than that under no-load downward operation. If analyzed in series without differentiation, normal fluctuations caused by changes in operating conditions may be misjudged as equipment degradation, or the true degradation trend may be obscured.

[0004] Some solutions can utilize PLC signals to obtain a unified time base, extract steady-state output segments, and generate working condition tags containing information such as load, speed, and position, which can be used for subsequent condition assessment or fault diagnosis. However, if the working conditions are not rigorously screened and corrected during the long-term degradation modeling stage, the obtained health evolution path will still be difficult to accurately reflect the actual degradation process of the target mechanism, easily leading to problems such as inaccurate identification of change points, degradation model drift, and large deviations in remaining life estimation. In addition, the instantaneous impact events unique to port cranes (such as spreader landing, brake action, and passing through rail joints) can introduce non-degradable signal jumps, which are difficult to effectively eliminate using conventional statistical outlier detection. For complex linkage operation modes of mechanisms such as trolleys and hoists, a single degradation model is also difficult to adapt to the degradation differences under different modes.

[0005] While existing technologies include a general health management process of "same working condition screening - change point detection - degradation modeling - life prediction," they lack specific and implementable strict rules and algorithms for port crane scenarios. For example, they do not propose strict judgment rules and quantitative distance calculation methods for the same and similar working conditions, do not provide regression correction methods to remove working condition disturbances, do not establish a mechanism to identify and remove or downweight abnormal samples associated with port crane operation events, do not solve the problem of multi-working-condition adaptive partitioning modeling, do not bind change point detection results with degradation model selection, and do not provide a calculation method for converting the remaining life from the number of comparable windows under the same working condition to the time scale. Therefore, they are difficult to directly apply to the online predictive maintenance of port cranes. Summary of the Invention

[0006] This invention addresses several issues in port crane mechanism condition monitoring, including significant interference from multiple operating conditions, a lack of rigorous judgment rules for constructing comparable sequences under the same operating conditions, a lack of reproducible operating condition correction and anomaly sample handling mechanisms for anchoring port crane operation events, a lack of binding relationship between change point detection and degradation modeling, and a lack of a method for calculating remaining life by mapping the number of comparable windows to a time scale. The invention proposes a method and system for port crane mechanism degradation modeling and remaining life estimation. This method constructs comparable feature sequences under the same operating conditions by building judgment rules for similar and identical operating conditions. It performs operating condition correction and removes or reduces the weight of non-degradation anomaly samples based on a healthy reference subset. It employs a degradation prediction approach that combines multi-operating-condition partitioning modeling with change point type binding. Finally, it uses the arrival rate of windows under the same or similar operating conditions to convert the number of remaining comparable windows into remaining life on a time scale, thereby improving the reliability of long-term degradation identification and predictive maintenance decisions for port crane mechanisms.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for modeling the degradation of port crane mechanisms and estimating their remaining life includes the following steps: Step A: Obtain the status monitoring feature value of the target mechanism or target mechanism component on the identified steady-state output segment, as well as the working condition tag and index information corresponding to the steady-state output segment; Step B: For the target mechanism and target working condition, according to the judgment rules of the same working condition and similar working condition, select the same working condition sample from the identified steady-state output segment, and supplement the similar working condition sample when the same working condition sample is insufficient. Combine at least one factor among steady-state output segment quality, sensor availability and data integrity for weighting and screening to construct a comparable feature sequence of the same working condition. Step C: Perform working condition correction on the comparable feature sequences under the same working condition based on a healthy reference subset, and remove or reduce the weight of non-degrading abnormal samples by combining the port machinery operation event anchor points to obtain the degradation modeling sequence; when the target mechanism has multiple operating modes or the sequence dispersion under a single working condition partition exceeds the preset threshold, perform multi-working condition partition modeling. Step D: Perform joint change point detection on the degradation modeling sequence or partition degradation modeling sequence to determine the valid change points and their change point types; Step E: Based on the type of change point, bind the corresponding degradation trajectory fitting model or future state prediction model to generate the future health evolution path of the target mechanism or target mechanism component; Step F: Determine the number of remaining comparable windows based on the future health evolution path and preset failure criteria, and convert the number of remaining comparable windows into an estimated remaining lifetime based on the arrival rate of windows with the same or similar operating conditions obtained from recent work log statistics. Step G: Output a strategy signal based on the estimated remaining lifetime.

[0008] On the other hand, the present invention also provides a port crane mechanism degradation modeling and remaining life estimation system for performing the above-mentioned port crane mechanism degradation modeling and remaining life estimation method. The system includes a data interface, processor, and memory that are communicatively connected to a PLC, condition monitoring sensors, and an operation log system, as well as functional modules called by the processor. The functional modules include: The data acquisition module is used to acquire the status monitoring characteristic values ​​of the target mechanism or its components in the identified steady-state output segment and their corresponding operating condition tags; The same working condition determination and sequence construction module is used to select samples of the same working condition according to the same working condition and similar working condition determination rules, and supplement samples of similar working conditions when they are insufficient. It combines the sample quality factor for weighting and screening to construct comparable feature sequences of the same working condition. The working condition correction and abnormal sample processing module is used to perform working condition correction based on the healthy reference subset, and to remove or downweight non-degraded abnormal samples in combination with the port machinery operation event anchor points. The multi-condition partition modeling module is used to perform multi-condition partition modeling when the target mechanism has multiple operating modes or the sequence dispersion exceeds a preset threshold. The joint change point detection module is used to perform joint change point detection on the degradation modeling sequence and determine the valid change points and their types. The model binding and degradation prediction module is used to bind degradation trajectory fitting models or future state prediction models based on the type of change point to generate future healthy evolution paths; The remaining lifetime output module is used to determine the estimated remaining lifetime value based on the failure criteria and the arrival rate of the same or similar operating condition window. The strategy generation module is used to output a strategy signal based on the estimated remaining lifetime.

[0009] Compared with the prior art, the beneficial effects of the present invention are: (1) By establishing specific rules for determining the same and similar working conditions for port cranes, and combining the degree of similarity of working conditions and sample quality to construct comparable feature sequences for the same working conditions, the interference of differences in load, speed, position, and linkage status on degradation trend analysis is reduced, and the comparability of degradation features is improved. In one embodiment, the above effect can be manifested as a reduction in the dispersion of the same working condition sequence, which can be verified by indicators such as the coefficient of variation, interquartile range, or absolute deviation of the median.

[0010] (2) By introducing a working condition correction mechanism based on a healthy reference subset and a mechanism for eliminating or downweighting abnormal samples associated with port machinery operation events, residual working condition disturbances can be removed, and pseudo-drifts caused by non-degenerative factors such as braking transients, courier impacts, and track joint impacts can be identified and processed, thereby improving the purity of the degradation modeling sequence. In one embodiment, the above effects can be manifested as a reduction in the correlation between condition monitoring features and load intervals, speed intervals, or position or angle intervals, as well as a reduction in the number of isolated jump samples caused by non-degenerative events such as braking switching, courier landing or departure, track joints, and limit neighborhoods.

[0011] (3) By adopting a multi-condition partitioning modeling method based on partition differences, partition fluctuations, effective coverage and sparse partitioning, it is possible to adaptively select whether to partition and how to partition in complex linkage scenarios such as lifting and trolley, thereby improving the adaptability of degradation modeling under different operating modes.

[0012] (4) By prioritizing the detection of change points and binding the degradation model according to the type of change points, the degradation initiation point and degradation acceleration interval can be identified more accurately, avoiding the mixing of healthy phase samples into the degradation fitting process, and improving the stability of change point localization and the reliability of life estimation. In historical playback samples with maintenance records, replacement records, or manually reviewed and marked intervals, the above effect can be verified by the number of deviation windows between the change point location and the abnormal development interval or the review and marked interval.

[0013] (5) By basing the remaining life calculation on the continuous failure criterion and the arrival rate of windows with the same or similar operating conditions, the degradation prediction results under comparable window scales can be converted into remaining life on a time scale, adapting to online predictive maintenance decisions under discrete operating conditions of port cranes. In one embodiment, the above effect can be verified by cross-period fluctuation indicators such as the standard deviation, range, or coefficient of variation of the remaining life estimate over multiple consecutive rolling evaluation periods.

[0014] (6) Based on the existing steady-state output segment and working condition tag, this invention has constructed a complete technical solution specifically for the long-term degradation identification and life management of mechanical components. It forms a clear and logically connected technical system architecture with the preceding steady-state extraction, state evaluation and fault attribution methods.

[0015] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0016] Figure 1 This is a schematic diagram illustrating the overall process and module relationships of the present invention; Figure 2 A schematic diagram of the process for constructing and backtracking indexes for comparable feature sequences under the same working conditions; Figure 3 A schematic diagram of the process for working condition correction, handling of port machinery operation event anchor point anomalies, and multi-working condition partition modeling. Figure 4 This is a schematic diagram illustrating the process of binding joint change point detection with degradation model; Figure 5 This diagram illustrates the remaining lifetime estimation and strategy signal output. Detailed Implementation

[0017] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings, so as to more clearly understand the purpose, features and advantages of this invention. It should be understood that the embodiments shown in the drawings are not intended to limit the scope of this invention, but are only for illustrating the essential spirit of the technical solutions of this invention. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0018] Unless the context requires otherwise, throughout the specification and claims, the word “comprising” and its variations, such as “including” and “having”, shall be understood to have an open, inclusive meaning, that is, to be interpreted as “including, but not limited to”.

[0019] Throughout this specification, references to "an embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Therefore, the appearance of "in an embodiment" or "an embodiment" in various places throughout the specification does not necessarily refer to the same embodiment. Furthermore, a particular feature, structure, or characteristic may be combined in any manner in one or more embodiments.

[0020] The singular forms “a” and “the” used in this specification and the appended claims include plural references unless otherwise expressly stated herein. It should be noted that the term “or” is generally used to mean “and / or” unless otherwise expressly stated herein.

[0021] In the following description, in order to clearly demonstrate the structure and working method of the present invention, a number of directional terms will be used. However, terms such as "front", "back", "left", "right", "outside", "inside", "outward", "inward", "up", and "down" should be understood as convenient terms and not as limiting terms.

[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments, but the scope of protection of the present invention is not limited to the following embodiments.

[0023] To facilitate understanding of the implementation boundaries of this invention, the following conventions are made for the core terms in this embodiment: (1) “Target mechanism” refers to the motion mechanism monitored and modeled in the port crane, including but not limited to hoisting mechanism, trolley traveling mechanism, gantry traveling mechanism, slewing mechanism and luffing mechanism; “Target mechanism component” refers to the component in the target mechanism that undertakes transmission, support, braking or driving functions, including but not limited to reducer, bearing, motor, drum, wire rope and brake.

[0024] (2) “Working condition tag (Tag)” refers to the set of working condition fields bound to a single steady-state output segment; “Target working condition tag” refers to the baseline working condition tag for which degradation modeling or life estimation is currently required. The working condition tag includes at least mechanism type, operation stage, load range, speed range and position or angle range, and may further include braking state, linkage state, directional state and environmental conditions.

[0025] (3) The “configuration version number” is used to identify the version differences of the target mechanism or target mechanism component before and after the replacement of major components, adjustment of sensor installation position, substantial change of sampling parameters or major mechanism modification; samples under the same configuration version are used to establish comparable degradation sequences, and samples under different configuration versions are not directly mixed into the same healthy reference subset in principle.

[0026] (4) "Health reference subset" refers to the data set that is not in an abnormal state after being confirmed by maintenance records, fault work orders, replacement records or manual review under the same mechanism type, the same operation stage and the same configuration version. It is used to estimate the target operating condition Tag reference center, operating condition disturbance components and health reference boundary.

[0027] (5) “Linkage Status” indicates whether the target mechanism is in a state of coordinated operation with other mechanisms; “Linkage Type” indicates the combination of participating mechanisms; “Linkage Intensity Level” indicates the speed ratio, duration ratio or other coordination level of the target mechanism and the cooperating mechanism in the steady-state output segment; “Linkage Status Change Record” indicates whether a linkage switching event has occurred in the steady-state output segment or its adjacent window.

[0028] (6) "Port crane operation event anchor point" refers to port crane operation events that can be determined by PLC, position encoder, limit switch, track joint position table, operation log or manually marked events, and is used to identify non-degenerate abnormal samples caused by brake switching, spreader loading or unloading, track joint passing, end limit neighboring area entry, linkage switching or sensor abnormality.

[0029] (7) "Number of remaining comparable windows" refers to the number of comparable windows remaining before the continuous failure criterion is met for the first time in the same or similar working condition windows after the current time; "Achievement rate of same or similar working condition windows" refers to the frequency of steady-state output segment windows that are the same or similar to the target working condition in the recent work log, which can be used to convert the degradation prediction results under the comparable window scale into the remaining lifetime estimate under the time scale.

[0030] This invention provides a method and system for degradation modeling and remaining life estimation of port crane mechanisms. The method takes the identified steady-state output segment and its corresponding operating condition tag as input. By constructing a comparable feature sequence of the same operating condition, and after operating condition correction, abnormal sample elimination or weight reduction and normalization based on port crane operation event anchor points, a degradation modeling sequence is obtained. On this basis, joint change point detection is performed and the degradation model is bound according to the change point type to generate a future healthy evolution path. Finally, the remaining life estimate and strategy signal are output based on the failure criterion and the arrival rate of the same or similar operating condition window.

[0031] like Figure 1 As shown, the overall process of the present invention includes: obtaining steady-state output segment feature values, operating condition tags, and related index information from the preceding process (step 101); performing same operating condition determination and constructing comparable sequences (step 102); performing operating condition correction, abnormal sample removal or weight reduction and normalization based on port machinery operation event anchor points (step 103); enabling multi-operating condition partitioning modeling when necessary (step 104); performing joint change point detection (step 105); performing degradation model binding and future prediction according to the change point type (step 106); and finally performing failure criterion judgment, remaining life estimation, and outputting strategy signals (step 107). The following is combined with... Figures 2 to 5 The key sub-processes in each step are explained in detail.

[0032] The following section uses the hoisting mechanism reducer and bearings of a quay crane (shore-mounted container crane) as the target object, and uses the effective value of vibration velocity, envelope spectrum sideband index, and temperature rise rate as condition monitoring characteristic values ​​to illustrate the complete implementation process of this invention. This invention is also applicable to the hoisting, trolley, crane, slewing, and luffing mechanisms of other types of port cranes (such as rail-mounted gantry cranes, tire-mounted gantry cranes, and portal cranes), requiring only the replacement of the corresponding speed, position, or angle fields and functional area definitions.

[0033] This embodiment uses the steady-state output segment and its corresponding operating condition tag obtained in the preceding process as input conditions. The preceding process has already completed unified time base alignment, steady-state window extraction, and the establishment of the input segment index; this invention does not repeat these steps. Each acquired sample contains a state monitoring feature value calculated on that steady-state output segment. And a working condition It is associated with a steady-state output segment identifier, an input segment index, a sensor identifier, and a configuration version number.

[0034] Step A: Obtain the identified steady-state output segment and its operating condition tag; The data acquisition and processing unit obtains the status monitoring characteristic values ​​of the target hoisting mechanism on the identified steady-state output segment, as well as the corresponding operating condition tag for the steady-state output segment. The characteristic values ​​include at least one or more of the following: effective vibration velocity, root mean square vibration acceleration, peak value, kurtosis, peak spectral value, envelope spectrum sideband index, cepstral index, temperature rise rate, oil particle concentration index, current characteristic value, or power characteristic value. The operating condition tag includes at least the mechanism type, operating stage, load range, speed range, and position or angle range, and may further include fields such as braking state, linkage state, directional state, and environmental conditions. The load range can be divided according to the rated load ratio or the load category of the operating command; the speed range can be divided according to the rated speed ratio; and the position or angle range can be divided in conjunction with the mechanism's stroke, functional area, or sensitive area.

[0035] Step B: Construct comparable feature sequences under the same working conditions; For the target mechanism (lifting) and the user-specified target working conditions (e.g., mechanism = lifting, operating phase = uniform speed ascent, load range = 50%~60% of rated load, speed range = 80%~90% of rated speed, position range = lifting height 20m~30m, braking state = stable release, linkage state = no linkage), execute the following sub-steps. The overall process of this step is as follows: Figure 2 As shown.

[0036] like Figure 2As shown, the construction of a comparable feature sequence under the same working condition includes: determining the target mechanism, target component, and target working condition (step 201); parsing the working condition tag and index field (step 202); determining whether candidate samples meet the hard constraints (step 203), and not including those that do not meet the hard constraints in the comparable sequence and recording the reasons (step 204); further determining whether samples that meet the hard constraints meet the sample quality threshold (step 205), and not including those that do not meet the hard constraints in the comparable sequence and recording the reasons (step 204); prioritizing the selection of samples under the same working condition from samples that meet the sample quality threshold (step 206), and determining the number of samples under the same working condition. If the requirements are met (step 207), then the sample weights are determined (step 211); if the requirements are not met, then the working condition distance is calculated and the soft distance is determined (step 208), and it is determined whether the similar working condition conditions are met (step 209). Those that do not meet the conditions are not included in the comparable sequence and the reason is recorded (step 204). Those that meet the conditions are supplemented as similar working condition samples and their source is marked (step 210). Then the sample weights are determined (step 211); finally, the samples are organized according to the time sequence and index information (step 212) to form a comparable feature sequence and backtracking index for the same working condition (step 213).

[0037] The specific execution steps are as follows: B1. Hard Constraint Filtering: First, filter out steady-state output segment samples that are completely consistent with the mechanism type, operation stage, and target working condition tag, where the braking state and linkage state have not switched, and the configuration version is the same. Segments where the braking state has switched or the linkage state type has changed are directly removed.

[0038] B2. Determine the degree of similarity between the working conditions: For each sample filtered through hard constraints, determine the degree of similarity between it and the target working condition tag in fields such as load range, speed range, position or angle range, direction status, and linkage strength level. For ordered fields such as load, speed, position, or angle, the degree of similarity can be determined by bin-level differences or preset proximity ranges; for discrete fields such as braking status, linkage type, and operation stage, they should be treated as hard constraints or exclusion conditions in advance.

[0039] In a preferred embodiment, the degree of proximity of operating conditions can be measured by the distance between operating conditions. express: in, , , They represent the first The binning differences or normalized differences between the individual samples and the target working condition tag in the load range, speed range, and position or angle range; It is an indicator function; Indicates differences in the intensity of linkage; , , , , These are preset weighting coefficients. The above formula is only one preferred implementation method for the working condition distance; in practical applications, the influencing factors can be reduced or adjusted according to the type of institution.

[0040] When samples meet the same operating conditions, they are preferentially included in the comparable feature sequence of the same operating conditions. When the number of samples with the same operating conditions is insufficient, samples with similar operating conditions that meet a preset threshold in terms of similarity and whose other items do not exceed the preset maximum deviation or have been weighted down by sample weights can be selected as supplementary samples. The threshold can be determined based on the distribution of the same and similar operating conditions in the health reference samples, historical misjudgments, or expert experience, and the source identifier of similar operating conditions is retained in the output results.

[0041] B3. Sample quality weighting and sequence generation: For each sample in the candidate set, the sample weight is determined by comprehensively considering factors such as the quality of the steady-state output segment, the availability of sensors and feature channels, similar operating conditions, and data integrity.

[0042] Samples that meet the quality threshold are retained, while samples with abnormal missing values, severe distortion, or non-steady-state conditions are removed. The filtered samples are sorted in chronological order, and each feature sample is associated with its steady-state output segment identifier, input segment index, sensor identifier, timestamp, and configuration version number to form a comparable feature sequence under the same operating conditions. A backtracking index is constructed to support subsequent tracing of the original waveform, input snapshots, and manual review.

[0043] Step C: Operating condition correction, abnormal sample handling, and normalization; The overall process of this step is as follows: Figure 3As shown. After inputting comparable feature sequences and backtracking indexes under the same operating conditions (step 301), the health reference subset and reference center are first extracted (step 302), and operating condition disturbance variables are extracted (step 303). Operating condition correction is performed based on the health reference subset (step 304). Then, the anchor points of port machinery operation events are identified (step 305), and non-degradation abnormal samples are removed or downweighted (step 306). On this basis, normalization and degradation direction unification are performed (step 307). When a single feature is insufficient to stably represent the degradation trend, it is determined whether to construct a combined health feature. If the index is positive (step 308), then perform multi-feature weighted fusion to form a combined health index (step 309), and then obtain the degradation modeling sequence (step 310); if not, then directly obtain the degradation modeling sequence (step 310); then determine whether multi-condition partitioning is needed (step 311); if needed, generate candidate partitioning schemes (step 312), perform partitioning scoring and sparse partitioning processing (step 313), and finally output the partition degradation modeling sequence (step 314); if not needed, then directly output the degradation modeling sequence (step 314).

[0044] The specific execution steps are as follows: C1. Selection of Healthy Reference Subset: For the target operating condition, a healthy reference subset is formed by selecting comparable samples from the most recent operating period that has been confirmed to be free of abnormalities and faults, has not involved major component replacements, sensor installation position adjustments, or substantial changes in sampling parameters. Preferably, the healthy reference subset prioritizes data from the same target mechanism or component within the same configuration version. When there are insufficient samples of the same target mechanism or component, data from similar mechanisms, with the same configuration version and traceable sources can be selected as supplementary data, while retaining the sample source identifier. The healthy reference subset does not include data from the post-maintenance break-in period, sensor replacement period, or major mechanism modification period, and should cover the main operating condition range required for subsequent modeling to ensure the stability of operating condition corrections.

[0045] C2. Operating Condition Correction: Based on the healthy reference subset, establish a relationship model between the original eigenvalues ​​and operating condition disturbance variables such as load deviation, speed deviation, position or angle deviation, environmental deviation, and linkage strength deviation, and use robust regression or local regression to estimate the degree of influence of each operating condition disturbance variable on the eigenvalues.

[0046] In a preferred embodiment, the relationship between the original state monitoring characteristic values ​​and the operating condition disturbance variables can be expressed as: in, For the first The original state monitoring feature values ​​of each sample; , , , , These represent the normalized deviations of load, velocity, position or angle, environment, and linkage intensity relative to the target working condition Tag reference center, respectively. As the benchmark, , , , , For regression coefficients, This is the residual term.

[0047] The eigenvalues ​​after working condition correction can be expressed as: in, The above processing represents the characteristic value after working condition correction; it means subtracting quantifiable working condition disturbance terms from the original characteristic value to reduce the impact of residual working condition differences such as load, speed, position or angle on the degradation trend.

[0048] Among them, load deviation, speed deviation, position or angle deviation, environmental deviation, and linkage strength deviation are all normalized and determined based on the reference center of the target working condition tag. The linkage strength deviation can be determined by the average speed ratio of the hoisting mechanism and the trolley in the corresponding steady-state output segment or the proportion of linkage duration; the environmental deviation can be deleted when environmental sensors are not deployed, and the remaining influencing factors can be re-estimated.

[0049] The corrected characteristic values ​​are obtained by subtracting the quantifiable estimate of the operating condition disturbance from the original condition monitoring characteristic values.

[0050] This correction can reduce the impact of quantifiable chemical condition disturbances on monitoring characteristics, allowing subsequent degradation modeling sequences to more primarily reflect changes in the mechanism's state.

[0051] C3. Abnormal Sample Removal or Weighting Based on Port Machinery Operation Event Anchor Points: The abnormal sample judgment result is comprehensively calculated based on the corrected samples. The abnormal sample judgment factors include the proximity of the port machinery operation event anchor point, the jump variable relative to the local median, the sample quality factor, the spatially sensitive area risk factor, and the signal quality penalty term.

[0052] Among them, the anchor points of port machinery operation events can be determined by one or more of the following: PLC brake status signal, spreader landing or landing signal, position encoder signal, limit switch signal, track joint position table, linkage status change record, or manually marked events. If the sample's corresponding steady-state output segment or its adjacent windows contain event flags such as brake release or holding, spreader landing or landing, passing through track joint, entering the end limit neighboring area, or linkage switching, the abnormality judgment level of the sample can be increased; if sensor saturation, data packet loss, flat lines, or severe distortion occur simultaneously, it can be directly removed or its weight can be significantly reduced.

[0053] C4. Normalization and Construction of Combined Health Index: The feature values ​​after removing or reducing the weight of outlier samples are normalized to obtain the degradation modeling sequence. When a single feature is insufficient to stably represent the degradation trend, a combined health index can be constructed based on multiple modified feature values.

[0054] The combined health index can be determined using a weighted fusion method of normalized features. The feature weights are set based on the target organization's degradation sensitivity, stability, and measurability, and can be adjusted through historical failure cases, maintenance records, or expert experience. The direction of the combined health index should be consistent: the larger the value, the higher the degree of degradation.

[0055] In a preferred embodiment, the combined health index can be expressed as: in, Indicates the first In the nth sample The normalized values ​​of the corrected features, Indicates the first The weights corresponding to each feature This indicates the number of features involved in the fusion.

[0056] C5. Multi-condition partitioning modeling: When the target mechanism has a complex operating mode and the sequence dispersion is still large under a single operating condition partition, multi-condition partitioning modeling is enabled. One or more candidate dimensions are selected from candidate dimensions such as load range, speed range, position or angle range, linkage state and environmental conditions to form candidate partitioning schemes, and each candidate partitioning scheme is comprehensively scored.

[0057] The comprehensive scoring factors include the degree of difference between partitions, residual fluctuation within partitions, effective coverage that meets the minimum sample size requirement, and the proportion of sparse partitions. The optimal partitioning scheme is selected based on the score meeting the preset selection criteria and the sample size of each partition meeting the lower limit requirement. If a sparse partition exists, the samples in that partition are merged into the partition closest to the operating condition that meets the sample size requirement; if multiple adjacent partitions exist, they are merged into the partition with the highest degree of similarity in operating conditions. The merged samples retain their original partition source identifier. After partitioning is determined, subsequent steps are performed on the corrected sequences within each partition. The degree of difference between partitions can be represented by the average Euclidean distance or inter-class variance between the centers of each partition; residual fluctuation within a partition can be represented by the weighted average of the coefficient of variation or variance within each partition; the effective coverage rate is the proportion of samples in partitions that meet the minimum sample size requirement out of the total sample size; the proportion of sparse partitions is the proportion of partitions with a sample size below the preset lower limit out of the total number of partitions.

[0058] Step D: Joint change point detection; Change point detection is performed on the degradation modeling sequence (or each partition sequence). The overall process of this step and subsequent degradation model binding is as follows: Figure 4 As shown. After inputting the degradation modeling sequence or the partition degradation modeling sequence (step 401), candidate change points and before-and-after comparison windows are set (step 402), and the mean difference, slope difference, fluctuation change, and continuous exceedance ratio are calculated (step 403), and then the joint change point score is calculated (step 404); then it is determined whether the effective change point conditions are met (step 405). If not, and a monotonic trend exists, the process enters the processing path for no significant change points detected but with a monotonic trend (step 406), and the whole-segment linear regression model is selected (step 412); if the effective change point conditions are met, the change is determined. The point type is determined (step 407), and the use of the sequences before and after the change point is separated (step 408). The sequence before the change point is used to determine the health reference interval and threshold boundary, and the sequence after the change point is used as the main interval for degradation modeling. Then, the corresponding model is selected according to the change point type (step 409). For drift or slow degradation types, a linear or piecewise linear model is selected (step 410). For accelerated degradation types, an exponential or probabilistic degradation model is selected (step 411). Finally, degradation trajectory fitting or future state prediction is performed (step 413) to generate the future health evolution path (step 414).

[0059] The specific execution steps are as follows: For each candidate change point in the degradation modeling sequence, the mean level, degradation slope, fluctuation amplitude, and continuous exceedance ratio of the sequence before and after the candidate point are compared, and the joint change point score is determined based on these factors.

[0060] In a preferred embodiment, candidate change points The joint change point score can be expressed as: in, This represents the change in mean before and after the candidate point of change. This represents the term indicating the change in the degradation slope. Indicates the fluctuation change term. This indicates the proportion of consecutive exceedances after a candidate change point. to These are preset weighting coefficients. The above scoring is used to illustrate a preferred implementation method for change point detection. In practical applications, the various items can be normalized or reduced according to the target institution and sample size.

[0061] In one exemplary calculation method, let the comparison window length before and after the candidate change point be... , Each point before and after the candidate change point can be taken as a candidate. The absolute difference of the sequence means within each window is normalized to the standard deviation of the health reference interval; It can be taken as the difference in the slope of the linear fitting within the previous and next windows; It can be taken as the ratio of the standard deviations of the front and back windows; After the point of change can be taken The proportion of samples that exceed the health reference boundary in each window.

[0062] Among them, the mean level is used to characterize whether the overall state index has risen or fallen; the degradation slope is used to characterize whether the degradation rate has changed; the fluctuation amplitude is used to characterize whether the state fluctuation has amplified; and the continuous exceedance ratio is used to characterize whether the sample continues to deviate from the healthy reference boundary after the change point. When the comprehensive score meets the preset change point judgment conditions and the number of consecutive valid samples after the change point... When the number of occurrences is not less than a preset number, the corresponding position is determined as a valid change point. When there are multiple change points in the sequence, repeated segmentation and detection can be performed.

[0063] The type of change point is determined based on the components of the joint change point score: If the mean change and slope change corresponding to the effective change point both meet the preset threshold, then the change point is determined to be the starting point of degradation; if the main manifestation is a slight shift in the mean but no obvious change in the slope, then it is determined to be the starting point of drift.

[0064] If a degradation initiation point already exists, and a new valid change point is detected in the subsequent sequence, and the degradation slope of the subsequent interval is significantly increased compared to the previous interval, then the change point is determined to be the starting point of the degradation acceleration interval.

[0065] Step E: Degradation model binding and future health evolution path generation; The sequence preceding the point of change is used to determine the healthy reference interval and threshold boundary, while the sequence following the point of change serves as the main interval for degradation modeling. The degradation model is selected based on the type of point of change: when the point of change corresponds to the start of drift, slow drift, or a slow degradation trend, a linear model or a piecewise linear model is used; when the point of change corresponds to accelerated degradation, an exponential model or a probabilistic degradation model is used. If no significant point of change is detected but the feature exhibits a monotonic trend, a whole-piece linear regression model is selected. Model parameters are determined through least-squares fitting or maximum likelihood estimation.

[0066] Using the established degradation model, the predicted values ​​of health indicators under multiple future operating conditions (same or similar) are extrapolated. This will generate pathways for future health evolution.

[0067] Step F: Remaining life estimation like Figure 5As shown, the process for remaining lifetime estimation and strategy signal output is as follows: input the future health evolution path (step 501), perform failure criterion judgment (step 502), and determine the number of remaining comparable windows N. rem (Step 503); Simultaneously input the recent work log (Step 504), and calculate and update the window arrival rate λ for the same or similar working conditions. tag (Step 505), and determine whether the arrival rate is sufficient to support the conversion (Step 506); if it is sufficient to support the conversion, directly perform RUL estimation (Step 508); if it is insufficient, indicate insufficient confidence or use conservative arrival rate estimation (Step 507), and then perform RUL estimation (Step 508); on this basis, perform risk level and confidence assessment (Step 509), generate strategy signal hierarchical output (Step 510), and output to maintenance terminal or equipment control system (Step 511). When it involves the limited operating condition signal executed by the control system, it should be effective after safety interlock or manual confirmation.

[0068] The specific execution steps are as follows: Failure criteria are defined, including at least one of the following: the combined health index reaches a preset failure threshold; the combined health index exceeds the normal boundary of the same working condition determined by the health reference subset for a consecutive preset number of windows with the same or similar operating conditions; when the combined health index is not constructed, a single modified state monitoring feature reaches its corresponding failure threshold; multiple modified features jointly satisfy their respective threshold conditions. Based on the predicted values ​​of the degradation model, starting from the windows with the same or similar operating conditions after the current time, it is determined whether the continuous exceedance condition is met one by one, and the position of the first occurrence of continuous exceedance is determined as the number of remaining comparable windows. The continuous exceedance refers to the situation where, in adjacent windows with the same or similar operating conditions in time sequence, the combined health index or the modified state monitoring feature value exceeds the health reference boundary determined by the health reference subset, and no window within the regression boundary appears in between.

[0069] The predicted health indicators for comparable future windows are extrapolated from the degradation model. The arrival rate of windows with the same or similar operating conditions is obtained by statistically analyzing the frequency of occurrence of corresponding operating condition windows in recent work records, with the unit being windows per day. The remaining comparable number of windows is converted to the window arrival rate to obtain the estimated remaining lifetime over the time scale. If the window arrival rate is insufficient to support the conversion, the output estimate confidence level is insufficient, or a preset conservative arrival rate is used for estimation, and the source of the conservative estimate is indicated in the output results.

[0070] In a preferred embodiment, the estimated remaining lifetime over a given time scale can be expressed as: in, This represents the number of remaining comparable windows that first meet the consecutive out-of-limit failure criterion starting from the current moment. This indicates the arrival rate of windows with the same or similar working conditions, obtained from recent work record statistics.

[0071] Arrival rates can be updated using a sliding time window or an index-weighted method. The length of the statistical window can be determined based on the port's operational scheduling cycle, equipment usage frequency, and sample size to reflect recent changes in operational pace.

[0072] Step G: Strategy signal output.

[0073] Based on the comparison between the estimated remaining lifespan and a preset safety threshold, a corresponding strategy signal is output. For example, when the estimated remaining lifespan is significantly higher than the preset warning threshold, a strategy signal is output to maintain the normal inspection cycle; when it approaches the warning threshold, a strategy signal is output to shorten the inspection cycle; when it is lower than the planned maintenance threshold, a strategy signal is output to arrange planned maintenance or limit load operation; and when it is lower than the shutdown threshold, a shutdown maintenance prompt signal is output. These strategy signals can be pushed to the maintenance terminal or sent to the equipment control system to trigger limited-condition operation.

[0074] When the control system executes a limited-condition operation command, it shall take effect after verification of the existing safety interlocks, authorization verification or manual confirmation of the port crane; the strategy signal is used to assist maintenance decision-making and operational risk control, and does not replace the original emergency stop, limit switch, anti-collision and safety protection logic of the equipment.

[0075] The above process takes the hoisting mechanism as an example. When applied to the trolley traveling mechanism, the position field is replaced with the trolley position range, the linkage status is changed to the linkage identifier with the hoisting mechanism, and the spatial sensitive area includes the front / rear limit neighborhood, seam area, etc. When applied to the slewing mechanism, the slewing angle range and angular velocity are used; the trolley traveling mechanism uses the traveling position range and traveling speed. The corresponding working distance, abnormal sample removal or weighting, change point detection, and failure criteria can all be implemented by reference, ensuring the comparability and interpretability of degradation identification and remaining life estimation under different mechanisms.

[0076] In the specific implementation process, the following key parameters can be determined through statistics, experience, or experimentation to ensure the feasibility of the solution. The relevant values ​​do not constitute a limitation on the scope of protection of this invention: Load, speed, and position are divided into compartments: functional areas are divided according to 10% of rated load, 5% to 10% of rated speed, and the entire stroke.

[0077] Threshold for similarity of working conditions: can be determined based on the distribution of the same and similar working conditions in the health sample set, historical misclassification, or expert experience.

[0078] Abnormal sample judgment threshold: It can be adjusted based on the score distribution of healthy samples, combined with the impact samples or abnormal sensor samples confirmed by maintenance.

[0079] Change point threshold: can be determined by replaying historical change point cases, taking into account the detection rate of change points, the false positive rate, and the number of valid samples after the change point.

[0080] Thresholds for the mean and slope terms can be determined by combining the statistical boundaries of the healthy reference subset, historical deterioration cases, and expert experience.

[0081] Failure threshold and number of consecutive out-of-limit windows: can be set based on historical failures, sample characteristic values ​​before replacement, maintenance records or expert experience.

[0082] Arrival rate statistics window length: can be determined based on port operation scheduling cycle, equipment operation frequency, and the frequency of occurrence of target operating conditions.

[0083] Robust estimation in regression correction can be achieved using iterative reweighted least squares, M-estimation, or local robust regression. Convergence conditions can be set according to sample size and engineering calculation requirements.

[0084] The parameters involved in this invention can be categorized into weighted parameters, regression estimation parameters, and judgment threshold parameters. Weighted parameters include those in the working condition distance formula. In the combined health index and in the joint change point score Preferably, the weight class parameters are all non-negative numbers and can be normalized to a certain value. Within the range, and the sum of the corresponding weights is 1.

[0085] In one implementation, the working condition distance weight satisfies... The combined health index weights meet the requirements. The joint change point scoring weights satisfy... Each weight can be determined based on the target institution's degradation sensitivity, stability within the healthy reference subset, missing rate, noise level, feature measurability, and the ability to distinguish historical samples.

[0086] For joint change point scoring, when the degradation of the target mechanism is mainly manifested as a trend of increasing, the weights of the mean change item and the slope change item can be increased; when the degradation is mainly manifested as increased shocks, amplified fluctuations, or continuous exceeding of limits, the weights of the fluctuation change item and the continuous exceeding of limits percentage item can be increased. The above weights can be determined through historical degradation case playback, manual review results, or expert experience, and can be recalibrated between different target mechanisms or target components.

[0087] Regression estimation parameters include the baseline term and regression coefficients in the operating condition correction model. These parameters are estimated from a healthy reference subset using least squares, iterative reweighted least squares, M-estimation, or local robust regression. Their values ​​are related to the dimensions of the condition monitoring characteristics, the normalization method, and the operating condition distribution of the target mechanism, and do not require a pre-fixed numerical range. When the input characteristic values ​​and operating condition disturbance variables have been normalized, the regression coefficients can be regularized, limited, or have outlier coefficients removed according to engineering needs.

[0088] Threshold parameters include minimum sample size threshold. Working condition distance threshold Length of the comparison window before and after the candidate change point Number of consecutive valid samples after the point of change Number of consecutive out-of-limit windows And the arrival rate statistics window length. As a preferred embodiment, 20 to 50 comparable windows can be selected. The distance distribution of the health reference subset can be set to 0.2 to 0.5 or the corresponding quantile. Five to twenty comparable windows can be selected. Five to 15 comparable windows can be selected. Two to five windows can be selected, and the length of the arrival rate statistics window can be 7 to 30 calendar days or the most recent 50 to 200 operation windows. The above values ​​are only preferred examples and can be adjusted according to the frequency of equipment use, the frequency of occurrence of the target operating condition, and the sample size.

[0089] In one implementation, the following parameter examples can be used to enhance the feasibility of the above steps; the listed values ​​are only for illustrating the parameter determination method and do not constitute a limitation on the scope of protection of this invention: In one parameter example, the minimum sample size threshold can be set according to the number of historical comparable windows for the target organization; when the sample size for the same operating condition is less than At that time, samples with similar operating conditions will be supplemented.

[0090] The working condition distance threshold can be determined based on the distance distribution of the same and similar working conditions in the health reference subset; in one implementation, it can be taken as... As a supplementary condition for similar operating conditions.

[0091] In change point detection, the length of the comparison window before and after a candidate change point can be taken as... For windows with the same or similar operating conditions, the number of consecutive valid samples after the point of change can be set to [value]. .

[0092] The number of consecutive out-of-limit windows in the failure criterion can be set according to the risk level of the target component; in one implementation, it can be taken as... That is, when three consecutive windows of the same or similar operating conditions meet the over-limit conditions, the life limit is determined to be reached.

[0093] In one computational flow-type embodiment, the hoisting mechanism reducer of the quay container crane is taken as the target mechanism component, and the target working condition tag is set as follows: the operation stage is uniform speed rise, the load range is 50% to 60% of the rated load, the speed range is 80% to 90% of the rated speed, the position range is the hoisting height of 20m to 30m, the braking state is stable release, the linkage state is no linkage, and the configuration version number is V1.

[0094] First, samples with the same mechanism type, operation stage, and configuration version number are selected from the preceding steady-state output segments, and no braking state switching or linkage type change has occurred. If there are 24 samples with the same working condition, which is less than the minimum sample number threshold, 12 samples are added from the similar working condition samples that meet the working condition distance threshold to form 36 candidate comparable windows, and the corresponding steady-state output segment identifier, input segment index, sensor identifier, and timestamp are retained.

[0095] Secondly, using the data from this configuration version that has been inspected and confirmed to be free of abnormalities and where no major components have been replaced as a healthy reference subset, the impact of load deviation, velocity deviation and position deviation on the target characteristics is estimated through robust regression. The condition disturbance component relative to the target condition tag reference center is subtracted from the original condition monitoring characteristic values ​​to obtain the corrected condition monitoring characteristic values.

[0096] Furthermore, if a comparable window is located within the neighboring window of an event such as brake release, sling landing or leaving the sling, track joint passage, entry into the end limit neighborhood, or linkage switching, and its jump variable relative to the local median exceeds a preset abnormal threshold, then the sample is deleted; if the jump variable is in the reduced weighting range, then the weight of the sample is reduced while the event anchor source identifier is retained.

[0097] Subsequently, the purified modified features are normalized and a degradation modeling sequence is formed. Ten identical or similar operating condition windows before and after the candidate change point are compared. When at least two of the mean difference, slope difference, fluctuation ratio, and continuous exceedance ratio meet the threshold, and the number of consecutive valid samples after the candidate change point is not less than 10, the candidate point is determined as a valid change point. The sequence before the change point is used to determine the healthy reference interval and threshold boundary, and the sequence after the change point is used as the main interval for degradation modeling.

[0098] Finally, if future health evolution paths indicate that the first consecutive exceedances from the current moment occur on the [missing information]th time... If there are multiple windows with the same or similar operating conditions, and these windows appear 35 times in the recent 14 days of work logs, then we can obtain... And thus obtain When the sample size within the arrival rate statistics window is too small or the arrival rate fluctuates beyond the preset range, the output estimate confidence level is insufficient, or a preset conservative arrival rate is used for conservative estimation, and the source of the conservative estimation is marked in the output results.

[0099] In some embodiments, the present invention also provides a port crane mechanism degradation modeling and remaining life estimation system, which is used to perform the steps described in the above method embodiments. The system includes a data interface, processor, and memory that are communicatively connected to a PLC, condition monitoring sensors, and an operation log system, as well as a data acquisition module, a same-condition determination and sequence construction module, a condition correction and abnormal sample processing module, a multi-condition partitioning modeling module, a joint change point detection module, a model binding and degradation prediction module, a remaining life output module, and a strategy generation module.

[0100] The data acquisition module is used to receive the status monitoring characteristic values ​​of the target mechanism or target mechanism component on the identified steady-state output segment and its corresponding working condition tag through the data interface. This module communicates with the data acquisition and processing unit, PLC, work log system, or status monitoring sensor, and receives the steady-state output segment characteristic values, working condition tag, steady-state output segment identifier, input segment index, sensor identifier, timestamp, and configuration version number from the preceding process output. The received working condition tag includes at least the mechanism type, operation stage, load range, speed range, and position or angle range, and may further include one or more of the following: braking state, linkage state, directional state, and environmental conditions.

[0101] The same working condition determination and sequence construction module is used to construct comparable feature sequences for the same working condition based on the rules for determining the same and similar working conditions. This module uses the target mechanism and target working condition as screening criteria. First, it performs hard constraint filtering to screen samples with consistent mechanism type, operating stage, configuration version, and no switching between braking and linkage states. Then, it determines the degree of similarity between each sample and the target working condition tag in terms of load range, speed range, position or angle range, direction state, and linkage intensity level. Finally, it combines sample quality factors and sensor availability factors to determine weights, organizes the samples chronologically to form comparable feature sequences for the same working condition, and constructs a backtracking index containing steady-state output segment identifiers, input segment indices, and timestamps.

[0102] The operating condition correction and abnormal sample processing module is used to perform healthy reference subset regression correction, abnormal sample elimination or weight reduction based on port machinery operation event anchor points, and normalization processing. This module first selects a healthy reference subset from comparable feature sequences under the same operating condition, and uses a robust regression method to establish the relationship between the original feature values ​​and load deviation, speed deviation, position or angle deviation, environmental deviation, and linkage strength deviation. Corrected feature values ​​are obtained by subtracting the operating condition disturbance estimate from the original feature values. Then, based on the proximity of port machinery operation event anchor points, local jump variables, spatially sensitive area risks, and signal quality penalties, abnormal samples caused by brake switching, spreader engagement or disengagement, track joint passage, entry into end-limit neighboring areas, linkage switching, and sensor distortion are identified, eliminated, or reduced. Finally, the purified sequence is normalized. When a single feature is insufficient to stably characterize the degradation trend, a combined health index is constructed.

[0103] The multi-condition partitioning modeling module is used to select a partitioning scheme and establish a partitioning degradation model when the dispersion of the degradation modeling sequence exceeds a preset threshold under multiple operating modes or a single operating condition partition. This module generates candidate partitioning schemes from candidate dimensions such as load range, speed range, position or angle range, linkage state, and environmental conditions. It comprehensively determines the partitioning score based on the difference between partitions, residual fluctuations within the partition, effective coverage, and the proportion of sparse partitions. Provided that the number of samples in each partition is not less than the minimum number of samples, the module selects the scheme that meets the preset optimization conditions as the optimal partitioning scheme. It also performs adjacent interval merging or similar operating condition migration to supplement sparse partitions, retaining the partitioning source identifier.

[0104] The joint change point detection module is used to identify degradation initiation points, drift initiation points, or degradation acceleration intervals. This module comprehensively compares the mean level, degradation slope, fluctuation amplitude, and continuous exceedance ratio before and after each candidate change point in the degradation modeling sequence. When the comprehensive judgment result meets the preset change point conditions and the number of consecutive valid samples after the change point meets the minimum sample number requirement, a valid change point is determined. The change point type is then determined based on the combined characteristics of mean change, slope change, and fluctuation change.

[0105] The model binding and degradation prediction module is used to perform degradation trajectory fitting or future state prediction based on the type of change point. This module uses the sequence before the change point to determine the health reference interval and threshold boundary, and the sequence after the change point as the main interval for degradation modeling. When the change point type corresponds to the drift start point, slow drift, or slow degradation trend, a linear model or a piecewise linear model is selected; when the change point type is degradation acceleration, an exponential model or a probabilistic degradation model is selected to generate the future healthy evolution path.

[0106] The remaining lifetime output module is used to output an estimated remaining lifetime value based on failure criteria and the arrival rate of windows under similar or identical operating conditions. This module calculates the number of remaining comparable windows that first meet the consecutive out-of-limit failure criteria from the current moment, based on the degradation model's predicted values. It then performs a ratio conversion based on the arrival rate of windows under similar or identical operating conditions obtained from recent operational records to obtain the estimated remaining lifetime value on the time scale. When the window arrival rate is insufficient to support the conversion, the output estimate has insufficient confidence or a preset conservative arrival rate is used for estimation.

[0107] The strategy generation module is used to output strategy signals based on the remaining life estimate. This module compares the remaining life estimate with preset first, second, and third safety thresholds, and outputs strategy signals for maintaining normal inspection cycles, shortening inspection cycles, scheduling planned maintenance, limiting load operation, or shutting down for maintenance, respectively. When the strategy signal is sent to the equipment control system and involves operating under limited conditions, it should take effect after safety interlock verification, authorization verification, or manual confirmation.

[0108] In terms of deployment, the same-condition determination and sequence construction module, the condition correction and abnormal sample processing module, and the joint change point detection module are deployed on edge computing gateways or industrial computers for real-time processing close to the data source. The multi-condition partitioning modeling module, model binding and degradation prediction module, remaining lifetime output module, and strategy generation module are deployed on edge servers, industrial servers, or cloud platforms to utilize stronger computing resources for model training, parameter updates, and long-term trend analysis. All modules communicate through standard data interfaces, and input data and output results are encapsulated in a unified format to ensure loose coupling and replaceability between system modules.

[0109] In some implementations, historical data from the hoisting mechanism of a quay crane can be used to verify the effectiveness of the invention. Comparison schemes may include: A) coarse screening based solely on mechanism and operational stage, without cascading tags for different operating conditions; B) introducing tags for the same or similar operating conditions to construct comparable sequences, but without performing operating condition correction or handling of outliers; C) adding operating condition correction and outlier removal or weighting based on B, but without partitioning modeling and binding change points; D) the complete scheme of the invention. Evaluation indicators may include the variation in dispersion of the same operating condition sequence, the correspondence between the location of change points and maintenance records, the misjudgment of change points, and the consistency between remaining life estimation and maintenance timing.

[0110] In one historical data playback embodiment, compared to scheme A, schemes B to D can progressively reduce the sequence dispersion caused by differences in load, speed, position, or angle. Specifically, when using the complete scheme D of this invention, based on the coefficient of variation, interquartile range, or median absolute deviation of the sequence corresponding to scheme A, the dispersion of the sequence under the same operating condition can be reduced by approximately 10%–30%. After correction by the healthy reference subset operating condition, the absolute value of the correlation between the condition monitoring characteristics and the load interval, speed interval, or position or angle interval can be reduced by approximately 15%–40%. After removing or downweighting the anchor points of port machinery operation events, the number of isolated jump samples caused by non-degenerative events such as braking switching, container engagement or disengagement, track joints, and limit neighborhoods can be reduced by approximately 20%–40%.

[0111] In samples with maintenance records, replacement records, or manually reviewed and marked intervals, the deviation between the identified change points and the corresponding abnormal development intervals or review and marked intervals can be controlled within approximately 2-6 windows of the same or similar operating conditions in one embodiment. Compared to the method of directly extrapolating at equal intervals according to natural time, the cross-cycle fluctuation of the remaining life estimate calculated based on the arrival rate of the same or similar operating condition windows can be reduced by approximately 10%-25%. The above quantitative range is used to illustrate the engineering improvement effect that the present invention can obtain under one or more historical data playback conditions, and does not constitute a limitation on the numerical value of the effect under specific equipment models, sample sizes, work intensity, or failure modes.

[0112] Among the above evaluation indicators, the dispersion of the sequence can be characterized by the coefficient of variation, interquartile range, or absolute deviation of the median; the correlation can be characterized by the absolute value of the Pearson correlation coefficient, Spearman correlation coefficient, or other preset correlation indicators between the condition monitoring characteristics and the load range, speed range, position, or angle range; isolated jump samples can be statistically analyzed based on the local median or the absolute deviation of the median threshold, combined with anchor points of port machinery operation events such as braking switching, container landing or departure, track joints, limit neighborhood, and linkage switching; cross-cycle fluctuations can be characterized by the standard deviation, range, or coefficient of variation of the remaining life estimate within the continuous rolling evaluation period.

[0113] Based on the same inventive concept, embodiments of the present invention also provide an electronic device, including a processor, a memory, and a communication interface. The memory stores a computer program, and when the computer program is executed by the processor, it implements the steps of the above-described port crane mechanism degradation modeling and remaining life estimation method.

[0114] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of the above-described port crane mechanism degradation modeling and remaining life estimation method.

[0115] Based on the same inventive concept, embodiments of the present invention also provide a computer program product, including program instructions, which, when executed by a processor, are used to implement the steps of the above-described port crane mechanism degradation modeling and remaining life estimation method.

[0116] Although the present invention has been described in detail with reference to the accompanying drawings and preferred embodiments, the invention is not limited thereto. Various equivalent modifications or substitutions can be made to the embodiments of the invention by those skilled in the art without departing from the spirit and essence of the invention. Such modifications or substitutions should all fall within the scope of the invention, or any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the invention should be covered within the protection scope of the invention. Therefore, the protection scope of the invention should be determined by the scope of the claims.

Claims

1. A method for modeling the degradation of port crane mechanisms and estimating their remaining life, characterized in that, Includes the following steps: Step A: Obtain the status monitoring feature value of the target mechanism or target mechanism component on the identified steady-state output segment, as well as the working condition tag and index information corresponding to the steady-state output segment; Step B: For the target mechanism and target working condition, according to the judgment rules of the same working condition and similar working condition, select the same working condition sample from the identified steady-state output segment, and supplement the similar working condition sample when the same working condition sample is insufficient. Combine at least one factor among steady-state output segment quality, sensor availability and data integrity for weighting and screening to construct a comparable feature sequence of the same working condition. Step C: Perform working condition correction on the comparable feature sequences under the same working conditions based on a healthy reference subset, and combine the port machinery operation event anchor points to remove or reduce the weight of non-degradation abnormal samples to obtain the degradation modeling sequence. When the target mechanism has multiple operating modes or the sequence dispersion under a single working condition partition exceeds a preset threshold, multi-working condition partition modeling is performed. Step D: Perform joint change point detection on the degradation modeling sequence or partition degradation modeling sequence to determine the valid change points and their change point types; Step E: Based on the type of change point, bind the corresponding degradation trajectory fitting model or future state prediction model to generate the future health evolution path of the target mechanism or target mechanism component; Step F: Determine the number of remaining comparable windows based on the future health evolution path and preset failure criteria, and convert the number of remaining comparable windows into an estimated remaining lifetime based on the arrival rate of windows with the same or similar operating conditions obtained from recent work log statistics. Step G: Output a strategy signal based on the estimated remaining lifetime; In step C, the working condition correction establishes a relationship model between the condition monitoring feature values ​​and the deviations of load, speed, position or angle, environment or linkage strength based on a healthy reference subset, and deducts the working condition disturbance component relative to the target working condition Tag reference center; the healthy reference subset is determined by data confirmed to be without anomalies under the same mechanism type, the same operation stage and the same configuration version; the port machinery operation event anchor point is determined by one or more of PLC signals, position or limit signals, track joint position table, operation log or manually marked events, and is used to remove or downweight abnormal samples caused by non-degradation events.

2. The method according to claim 1, characterized in that, The working condition tag in step A includes at least mechanism type, working stage, load-related fields, speed-related fields, and position or angle-related fields, and includes braking, linkage, direction, or environment-related fields; the index information includes steady-state output segment identifier, input segment index, sensor identifier, timestamp, or configuration version number; the target mechanism includes one or more of hoisting, trolley, crane, slewing, or luffing mechanisms; the target mechanism components include one or more of reducers, bearings, or motors; and the condition monitoring characteristic values ​​include one or more of vibration, temperature, oil, current, or power-related condition characteristics.

3. The method according to claim 1, characterized in that, The rules for determining the same and similar working conditions in step B include hard constraint filtering and soft distance determination. The hard constraint filtering requires that the mechanism type, operation stage, and configuration version be consistent, and excludes steady-state output segments where braking state switching or linkage type changes occur. The soft distance determination determines the degree of similarity of working conditions based on the differences between the load, speed, position or angle, direction, or linkage intensity and the target working condition tag. The same working condition sample is a sample whose load, speed, position, or angle-related working condition fields are consistent with the target working condition tag. The similar working condition sample is a sample whose working condition fields meet the preset proximity range and do not exceed the preset maximum deviation or have been processed by sample weight reduction.

4. The method according to claim 1, characterized in that, In step C, the multi-condition partition modeling selects a partition scheme based on the degree of difference between partitions, residual fluctuations within partitions, effective coverage, and the proportion of sparse partitions. For partitions with insufficient sample numbers, adjacent partitions are merged or similar conditions are migrated to supplement the sample. When a single feature is insufficient to characterize the degradation trend, multiple modified state monitoring features are normalized and weighted to construct a combined health index with a larger value indicating a higher degree of degradation.

5. The method according to claim 1, characterized in that, The joint change point detection in step D includes comparing the mean level, degradation slope, fluctuation amplitude, and the proportion of samples continuously exceeding the health reference boundary before and after the candidate change point to determine the joint change point score; when the joint change point score meets the preset change point judgment condition and the number of valid samples after the change point meets the preset requirement, the valid change point is determined, and the change point type is determined according to the mean change, slope change, fluctuation change, and continuous exceeding of the limit. The joint change point score is determined by the following formula: ； in, This represents the change in mean before and after the candidate point of change. This represents the term indicating the change in the degradation slope. Indicates the fluctuation change term. Indicates the proportion of consecutive exceedances after a candidate change point. to These are preset weighting coefficients.

6. The method according to claim 1 or 5, characterized in that, In step E, the sequence before the effective change point is used to determine the healthy reference interval and threshold boundary, and the sequence after the effective change point is used as the main interval for degradation modeling. When the change point type corresponds to drift or slow degradation, a linear model or a piecewise linear model is used; when the change point type corresponds to accelerated degradation, an exponential model or a probabilistic degradation model is used; when no effective change point is detected but the degradation modeling sequence shows a monotonic trend, a whole-segment linear regression model is used.

7. The method according to claim 1, characterized in that, The failure criteria in step F include the combined health index or a single corrected condition monitoring feature value reaching a preset failure threshold, a preset number of consecutive windows of the same or similar operating conditions exceeding the health reference boundary, or multiple corrected features jointly satisfying at least one of the failure conditions. The number of remaining comparable windows is determined by the position where the future health evolution path first satisfies the continuous over-limit failure criterion. The arrival rate of the same or similar operating condition windows is updated based on recent work log statistics using a sliding time window or exponential weighting method. The remaining comparable window number is divided by the arrival rate of the same or similar operating condition windows to obtain the estimated remaining lifetime value on the time scale. The continuous over-limit refers to the combined health index or corrected condition monitoring feature value continuously exceeding the health reference boundary in adjacent windows of the same or similar operating conditions in time sequence, without any intermediate regression to the boundary.

8. The method according to claim 1 or 7, characterized in that, The strategy signals mentioned in step G include inspection cycle adjustment, load-limited operation, shutdown maintenance reminder, or control system limited operating condition operation signals; among them, the limited operating condition operation signals involving the execution of the equipment control system take effect after safety interlock verification, authorization verification, or manual confirmation.

9. A system for modeling the degradation of port crane mechanisms and estimating their remaining life, characterized in that, A method for performing degradation modeling and remaining life estimation of a port crane mechanism as described in any one of claims 1 to 8, comprising a data interface, a processor, a memory, and a function module called by the processor, all connected to a PLC, condition monitoring sensors, and an operation log system, wherein the function module includes: The data acquisition module is used to acquire the status monitoring characteristic values ​​of the target mechanism or its components in the identified steady-state output segment and their corresponding operating condition tags; The same working condition determination and sequence construction module is used to select samples of the same working condition according to the same working condition and similar working condition determination rules, and supplement samples of similar working conditions when they are insufficient. It combines the sample quality factor for weighting and screening to construct comparable feature sequences of the same working condition. The working condition correction and abnormal sample processing module is used to perform working condition correction based on the healthy reference subset, and to remove or downweight non-degraded abnormal samples in combination with the port machinery operation event anchor points. The multi-condition partition modeling module is used to perform multi-condition partition modeling when the target mechanism has multiple operating modes or the sequence dispersion exceeds a preset threshold. The joint change point detection module is used to perform joint change point detection on the degradation modeling sequence and determine the valid change points and their types. The model binding and degradation prediction module is used to bind degradation trajectory fitting models or future state prediction models based on the type of change point to generate future healthy evolution paths; The remaining lifetime output module is used to determine the estimated remaining lifetime value based on the failure criteria and the arrival rate of the same or similar operating condition window. The strategy generation module is used to output a strategy signal based on the estimated remaining lifetime.

Images