Industrial robot fault prediction and health management system based on digital twinning

Abstract

The invention discloses an industrial robot fault prediction and health management system based on digital twinning, and relates to the technical field of robot fault monitoring. According to the method, joint current data, joint temperature data, following error data, alarm event data, program segment number data and maintenance action data in an industrial robot controller and a maintenance terminal are sorted into an operation data set, a maintenance record set and a working condition label set, and then a prediction root cause set is obtained through a generated twin baseline set; the pre-maintenance parameter set and the post-maintenance parameter set are obtained, and finally, the pre-maintenance parameter set and the post-maintenance parameter set are combined with the read preset root cause association relationship and the preset maintenance strategy rule to generate a fake result set, and the knowledge set and the health conclusion set are updated, so that surface judgment is not made only according to whether an alarm disappears or not or whether equipment resumes operation or not. However, the judgment result before maintenance can be verified again, and the verification result is continuously fed back to the subsequent judgment link.

Description

Technical Field

[0001] This invention relates to the field of robot fault monitoring technology, specifically to an industrial robot fault prediction and health management system based on digital twins. Background Technology

[0002] As industrial automated production lines develop towards greater flexibility, continuity, and high cycle times, industrial robots have been widely applied in production processes such as automotive welding, parts handling, assembly, grinding, painting, and inspection. During long-term operation, industrial robots not only perform repetitive motion tasks but also continuously withstand the effects of load changes, cycle time fluctuations, heat accumulation, and process switching. Therefore, the operating status of industrial robots is closely related to the stability of the production line cycle time, product consistency, and equipment maintenance costs.

[0003] In existing industrial settings, after receiving early warning or fault diagnosis results, maintenance personnel typically perform repairs based on experience rules, alarm information, or recommended work orders. These repairs may include replacing reducers, adjusting lubrication, correcting control parameters, or handling servo anomalies. After repairs are completed, the system usually judges the effectiveness of the repair based solely on whether the alarm has disappeared, whether the equipment has been restarted, and whether it has been shut down again within a short period. However, it lacks a consistency verification mechanism between the predicted root causes before repairs and the actual parameter changes after repairs. In other words, while existing systems can provide judgments before repairs, they cannot verify whether the preceding judgments hit the actual degradation source after repairs through parameter inversion. Nor can they write the verification results back into the health management model and fault causal relationships, resulting in a long-term disconnect between the fault prediction link and the repair execution link. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an industrial robot fault prediction and health management system based on digital twins, which solves the problems mentioned in the background technology.

[0005] The present invention is achieved through the following technical solution: a fault prediction and health management system for industrial robots based on digital twins, including a data processing module, a prediction modeling module, a parameter inversion module, and a falsification and write-back module;

[0006] The data processing module collects joint current data Jcur, joint temperature data Jtmp, following error data Ferr, alarm event data Almr, program segment number data Segm, and maintenance action data Mact from the industrial robot controller and maintenance terminal, and performs time alignment and working condition classification to form the running dataset Datr, the maintenance record set Drec, and the working condition label set Tagn.

[0007] The predictive modeling module extracts baseline segments from the running dataset Datr based on the working condition label set Tagn and the maintenance interval Mwin, generating a twin baseline set Bsim; and extracts the pre-maintenance segment set Sbef from the running dataset Datr based on the working condition label set Tagn and the maintenance interval Mwin, and compares the pre-maintenance segment set Sbef with the twin baseline set Bsim to obtain the predicted root cause set Caus.

[0008] The parameter inversion module performs parameter inversion on the pre-maintenance and post-maintenance operation segments based on the twin baseline set Bsim, obtaining the pre-maintenance parameter set Pbef and the post-maintenance parameter set Paft.

[0009] The falsification and write-back module reads the preset root cause correlation and preset maintenance strategy rules, and performs consistency judgment based on the predicted root cause set Caus, the pre-maintenance parameter set Pbef, the post-maintenance parameter set Paft, and the maintenance record set Drec to generate a falsification result set Fals; then, it modifies the read preset root cause correlation and preset maintenance strategy rules according to the falsification result set Fals to form an updated knowledge set Know; finally, it outputs a health conclusion set Heal based on the updated knowledge set Know and the post-maintenance parameter set Paft.

[0010] Preferably, the data processing module includes a data access unit;

[0011] The data access unit reads joint current data Jcur, joint temperature data Jtmp, following error data Ferr, alarm event data Almr, and program segment number data Segm from the industrial robot controller, and reads maintenance action data Mact from the maintenance terminal.

[0012] Then, establish associations based on the same industrial robot, the same program execution process, and the same maintenance process, and collect the reading results in a unified manner to form the original association set Raws.

[0013] Preferably, the data processing module further includes a time-series classification unit;

[0014] The time series classification unit divides the operating condition interval Cseg based on the original association set Raws and according to the change order of the program segment number data Segm on the time axis.

[0015] Align the joint current data Jcur, joint temperature data Jtmp, follow-up error data Ferr, and alarm event data Almr, which fall within the same operating condition range Cseg, to the corresponding program segment number data Segm.

[0016] Then, the aligned joint current data Jcur, joint temperature data Jtmp, follow error data Ferr, alarm event data Almr, and corresponding program segment number data Segm are written into the runtime dataset Datr.

[0017] The maintenance interval Mwin is determined by the occurrence time of the first maintenance action data Mact and the last maintenance action data Mact in the same maintenance process, and the maintenance action data Mact located in the maintenance interval Mwin and the corresponding program segment number data Segm are written into the maintenance record set Drec.

[0018] Then, based on the program segment number data Segm corresponding to each working condition interval Cseg and the position of each working condition interval Cseg relative to the maintenance interval Mwin, a working condition label set Tagn is generated.

[0019] Preferably, the predictive modeling module includes a baseline generation unit;

[0020] The baseline generation unit filters the labels whose corresponding working condition interval Cseg is located before the maintenance interval Mwin based on the working condition label set Tagn, and arranges the filtered labels in chronological order.

[0021] Then, remove a preset number of tags that are closest to the maintenance interval Mwin in terms of time sequence, and form the baseline tag set Btag with the remaining tags;

[0022] Extract the joint current data Jcur, joint temperature data Jtmp, follow error data Ferr, alarm event data Almr, and program segment number data Segm corresponding to the baseline tag set Btag from the running dataset Datr to form the baseline segment set Bref;

[0023] The data in the baseline segment set Bref that have the same working condition label set Tagn and the same program segment number data Segm are matched according to the time order. The average values ​​of the joint current data Jcur, joint temperature data Jtmp and follow-up error data Ferr at the corresponding positions are calculated to form the current reference set Curb, the temperature rise reference set Tmpb and the error reference set Errb.

[0024] Then, write the current reference set Curb, temperature rise reference set Tmpb, and error reference set Errb corresponding to each operating condition label set Tagn into the twin baseline set Bsim.

[0025] Preferably, the predictive modeling module further includes a root cause extraction unit;

[0026] The root cause extraction unit filters a preset number of tags whose corresponding working condition interval Cseg is located before the maintenance interval Mwin and is closest to the maintenance interval Mwin in time order, based on the working condition tag set Tagn and maintenance interval Mwin, to form a preceding tag set Ptag.

[0027] Extract the joint current data Jcur, joint temperature data Jtmp, follow-up error data Ferr, alarm event data Almr, and program segment number data Segm from the running dataset Datr, which correspond to the preceding tag set Ptag, to form the pre-maintenance segment set Sbef.

[0028] Extract the current reference set Curb, temperature rise reference set Tmpb, and error reference set Errb corresponding to the preceding tag set Ptag from the twin baseline set Bsim;

[0029] Subtract the joint current data Jcur in the pre-maintenance segment set Sbef from the corresponding current reference set Curb item by item according to the same program segment number data Segm and the same order position to form the current difference set Dcur.

[0030] Subtract the joint temperature data Jtmp in the pre-maintenance segment set Sbef from the corresponding temperature rise reference set Tmpb by subtracting the data with the same program segment number Segm and the same order of position to form the temperature rise difference set Dtmp.

[0031] Subtract the following error data Ferr in the pre-maintenance segment set Sbef from the corresponding error reference set Errb item by item according to the same program segment number data Segm and the same order position to form the error difference set Derr.

[0032] The current difference set Dcur, the temperature rise difference set Dtmp, and the error difference set Derr are combined according to the corresponding preceding label set Ptag, the corresponding program segment number data Segm, and the corresponding sequence position to form the deviation relationship set Devr;

[0033] Based on the deviation relationship set Devr and the corresponding alarm event data Almr, generate the hysteresis judgment value Bres, friction judgment value Fres, thermal drift judgment value Tres and delay judgment value Lres respectively.

[0034] Write the hysteresis judgment value Bres, friction judgment value Fres, thermal drift judgment value Tres, and delay judgment value Lres into the prediction root cause set Caus according to the corresponding preceding label set Ptag and the corresponding program segment number data Segm.

[0035] Preferably, the parameter inversion module includes a post-extraction unit;

[0036] The subsequent extraction unit, based on the working condition label set Tagn, the maintenance interval Mwin, and the preceding label set Ptag, filters labels whose corresponding working condition interval Cseg is located after the maintenance interval Mwin and whose working condition label is the same as the working condition label in the preceding label set Ptag, to form the subsequent label set Atag.

[0037] Then, extract the joint current data Jcur, joint temperature data Jtmp, follow-up error data Ferr, alarm event data Almr, and program segment number data Segm from the running dataset Datr to form the post-maintenance segment set Saft.

[0038] Preferably, the parameter inversion module further includes a pre-parameter solving unit;

[0039] The pre-parameter solving unit extracts the current reference set Curb, temperature rise reference set Tmpb, and error reference set Errb corresponding to the pre-label set Ptag from the twin baseline set Bsim, forming the pre-reference segment set Bpre;

[0040] Subtract the joint current data Jcur in the pre-maintenance segment set Sbef from the current reference set Curb in the pre-maintenance reference segment set Bpre by subtracting each item from the data with the same program segment number Segm and the same order position to form the pre-current difference set Cpre.

[0041] Subtract the joint temperature data Jtmp in the pre-maintenance segment set Sbef from the temperature rise reference set Tmpb in the pre-maintenance reference segment set Bpre by subtracting each item from the data with the same program segment number Segm and the same order position to form the pre-temperature rise difference set Tpre.

[0042] Subtract the following error data Ferrr in the pre-maintenance segment set Sbef from the error reference set Errb in the pre-maintenance reference segment set Bpre by subtracting each item from the data with the same program segment number Segm and the same order position to form the pre-error difference set Epre.

[0043] Based on the alarm event data Almr corresponding to the previous current difference set Cpre, previous temperature rise difference set Tpre, previous error difference set Epre and previous maintenance segment set Sbef, the previous backlash value Bvbf, previous friction value Fvbf, previous thermal drift value Tvbf and previous delay value Lvbf are generated respectively.

[0044] Simultaneously, based on the previous backlash value Bvbf, previous friction value Fvbf, previous thermal drift value Tvbf, and previous delay value Lvbf, the backlash parameter Back, friction parameter Fric, thermal drift parameter Ther, and delay parameter Dely in the pre-maintenance parameter set Pbef are determined respectively, thus forming the pre-maintenance parameter set Pbef.

[0045] Preferably, the parameter inversion module further includes a post-parameter solving unit;

[0046] The post-parameter solving unit extracts the current reference set Curb, the temperature rise reference set Tmpb, and the error reference set Errb corresponding to the subsequent tag set Atag from the twin baseline set Bsim, forming the post-reference segment set Baft;

[0047] Then, subtract the joint current data Jcur in the post-repair segment set Saft from the current reference set Curb in the post-repair reference segment set Baft item by item according to the same program segment number data Segm and the same order position to form the post-current difference set Caft.

[0048] Subtract the joint temperature data Jtmp in the post-maintenance segment set Saft from the temperature rise reference set Tmpb in the post-base segment set Baft, using the same program segment number data Segm and the same order position, to form the post-temperature rise difference set Taft.

[0049] Subtract the follow-up error data Ferrr in the post-maintenance segment set Saft from the error reference set Errb in the post-basic segment set Baft, item by item according to the same program segment number data Segm and the same order position, to form the post-error difference set Eaft;

[0050] Based on the alarm event data Almr corresponding to the post-current difference set Caft, post-temperature rise difference set Taft, post-error difference set Eaft, and post-maintenance segment set Saft, the post-hysteresis value Bvaf, post-friction value Fvaf, post-thermal drift value Tvaf, and post-delay value Lvaf are generated respectively.

[0051] Simultaneously, based on the back back value Bvaf, back friction value Fvaf, back thermal drift value Tvaf, and back delay value Lvaf, the back back parameter Back, friction parameter Fric, thermal drift parameter Ther, and delay parameter Dely in the post-maintenance parameter set Paft are determined respectively, thus forming the post-maintenance parameter set Paft.

[0052] Preferably, the falsification write-back module includes a consistency determination unit;

[0053] The consistency determination unit extracts the maintenance action data Mact corresponding to the current maintenance process from the maintenance record set Drec, and determines the maintenance object set Mobj based on the maintenance action data Mact;

[0054] Subtract the hysteresis parameter Back in the parameter set Pbef before maintenance from the hysteresis parameter Back in the parameter set Paft after maintenance to obtain the hysteresis change value Bchg;

[0055] The friction parameter Fric in the parameter set Pbef before maintenance is subtracted from the friction parameter Fric in the parameter set Paft after maintenance to form the friction change value Fchg;

[0056] Subtract the thermal drift parameter Ther in the parameter set Pbef before maintenance from the thermal drift parameter Ther in the parameter set Paft after maintenance to obtain the thermal drift change value Tchg;

[0057] Subtract the delay parameter Dely in the parameter set Pbef before maintenance from the delay parameter Dely in the parameter set Paft after maintenance to obtain the delay change value Lchg;

[0058] Write the hysteresis change value Bchg, friction change value Fchg, thermal drift change value Tchg, and delay change value Lchg into the parameter change set Vary;

[0059] Then, the predicted root cause set Caus, the maintenance object set Mobj, and the parameter change set Vary are compared item by item, and the consistent or inconsistent results obtained from each comparison are written into the consistency result set Cons.

[0060] Then, extract the inconsistent results from the consistent result set Cons and write the extracted inconsistent results into the falsification result set Fals.

[0061] Preferably, the falsification write-back module further includes a write-back output unit;

[0062] The write-back output unit reads the preset root cause correlation and preset maintenance strategy rules;

[0063] Then, based on the falsification result set Fals, the root cause association items corresponding to the inconsistent results are determined from the read preset root cause association relationships, and the determined root cause association items are corrected.

[0064] At the same time, based on the falsification result set Fals, the maintenance strategy items corresponding to the inconsistent results are determined from the read preset maintenance strategy rules, and the determined maintenance strategy items are corrected.

[0065] Then, write the revised root cause correlation items and revised maintenance strategy items into the updated knowledge set Know;

[0066] Then, based on the updated knowledge set Know and the post-repair parameter set Paft, the health conclusion set Heal is output.

[0067] This invention provides a fault prediction and health management system for industrial robots based on digital twins, which has the following beneficial effects:

[0068] (1) First, the joint current data Jcur, joint temperature data Jtmp, following error data Ferr, alarm event data Almr, program segment number data Segm, and maintenance action data Mact in the industrial robot controller and maintenance terminal are organized into a running dataset Datr, a maintenance record set Drec, and a working condition label set Tagn. Then, the predictive modeling module generates a twin baseline set Bsim and obtains a predictive root cause set Caus. Subsequently, the parameter inversion module obtains the pre-maintenance parameter set Pbef and the post-maintenance parameter set Paft. Finally, the falsification write-back module combines the read preset root cause correlation relationship and preset maintenance strategy rules to generate a falsification result set Fals, an updated knowledge set Know, and a health conclusion set Heal. This enables the system to no longer just make a superficial judgment based on whether the alarm disappears or whether the equipment resumes operation, but to re-verify the judgment result before maintenance and continue to feed the verification result back to the subsequent judgment link.

[0069] (2) Extract the pre-maintenance segment set Sbef from the preceding tag set Ptag that is closest to the maintenance interval Mwin in time sequence, and compare the joint current data Jcur, joint temperature data Jtmp and following error data Ferr with the corresponding benchmark results in the twin baseline set Bsim item by item to form the current difference set Dcur, temperature rise difference set Dtmp, error difference set Derr and deviation relationship set Devr. Then, obtain the hysteresis judgment value Bres, friction judgment value Fres, thermal drift judgment value Tres and delay judgment value Lres and write them into the prediction root cause set Caus. Instead of drawing conclusions directly based on a certain alarm or a certain instantaneous parameter, quantitatively compare the abnormal changes before the maintenance under the same working condition benchmark reference, so as to distinguish the normal load changes and temperature rise changes of the program segment itself from the real abnormal deterioration.

[0070] (3) Based on the twin baseline set Bsim, the pre-reference segment set Bpre and the post-reference segment set Baft are formed, and the pre-current difference set Cpre, the pre-temperature rise difference set Tpre, the pre-error difference set Epre, the post-current difference set Caft, the post-temperature rise difference set Taft, and the post-error difference set Eaft are further obtained. Finally, the pre-maintenance parameter set Pbef and the post-maintenance parameter set Paft are formed respectively, thereby transforming the pre-maintenance anomaly and post-maintenance recovery from phenomenon judgment to directly comparable structured parameter results, such as the changes before and after the hysteresis parameter Back, the friction parameter Fric, the thermal drift parameter Ther, and the delay parameter Dely, so as to quantify whether the improvement before and after maintenance is real, and avoid misjudging the maintenance effectiveness on site based solely on whether the alarm event data Almr has disappeared or whether the equipment has temporarily resumed operation. Attached Figure Description

[0071] Figure 1This is a diagram illustrating the overall structure of a digital twin-based industrial robot fault prediction and health management system.

[0072] Figure 2 A schematic diagram of the data processing module structure;

[0073] Figure 3 A schematic diagram of the predictive modeling module structure;

[0074] Figure 4 This is a schematic diagram of the parameter inversion module structure;

[0075] Figure 5 This is a schematic diagram of the falsification write-back module structure. Detailed Implementation

[0076] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings in the embodiments of the present invention.

[0077] Example 1

[0078] This invention provides a fault prediction and health management system for industrial robots based on digital twins. Please refer to [link / reference]. Figures 1 to 5 It includes a data processing module, a predictive modeling module, a parameter inversion module, and a falsification and write-back module;

[0079] The data processing module collects joint current data Jcur, joint temperature data Jtmp, following error data Ferr, alarm event data Almr, program segment number data Segm, and maintenance action data Mact from the industrial robot controller and maintenance terminal, and performs time alignment and working condition classification to form the running dataset Datr, the maintenance record set Drec, and the working condition label set Tagn.

[0080] The predictive modeling module extracts baseline segments from the running dataset Datr based on the working condition label set Tagn and the maintenance interval Mwin, generating a twin baseline set Bsim; and extracts the pre-maintenance segment set Sbef from the running dataset Datr based on the working condition label set Tagn and the maintenance interval Mwin, and compares the pre-maintenance segment set Sbef with the twin baseline set Bsim to obtain the predicted root cause set Caus.

[0081] The parameter inversion module performs parameter inversion on the pre-maintenance and post-maintenance operation segments based on the twin baseline set Bsim, obtaining the pre-maintenance parameter set Pbef and the post-maintenance parameter set Paft.

[0082] The falsification and write-back module reads the preset root cause correlation and preset maintenance strategy rules, and performs consistency judgment based on the predicted root cause set Caus, the pre-maintenance parameter set Pbef, the post-maintenance parameter set Paft, and the maintenance record set Drec to generate a falsification result set Fals; then, it modifies the read preset root cause correlation and preset maintenance strategy rules according to the falsification result set Fals to form an updated knowledge set Know; finally, it outputs a health conclusion set Heal based on the updated knowledge set Know and the post-maintenance parameter set Paft.

[0083] In this embodiment, the data processing module first organizes the joint current data Jcur, joint temperature data Jtmp, following error data Ferr, alarm event data Almr, program segment number data Segm, and maintenance action data Mact from the industrial robot controller and maintenance terminal into a running dataset Datr, a maintenance record set Drec, and a working condition label set Tagn. Then, the predictive modeling module generates a twin baseline set Bsim and obtains a predicted root cause set Caus. Subsequently, the parameter inversion module obtains the pre-maintenance parameter set Pbef and the post-maintenance parameter set Paft. Finally, the falsification and write-back module combines the read preset root cause correlation and preset maintenance strategy rules to generate a falsification result set Fals, an updated knowledge set Know, and a health conclusion set Heal. This allows the system to move beyond superficial judgments such as "whether the alarm has disappeared" or "whether the equipment has resumed operation" and instead re-verify whether the pre-maintenance judgment hit the real degradation source, and continue to feed the verification results back to subsequent judgment links. For example, in an automotive body-in-white welding production line, an industrial robot is judged to have friction abnormality by the predicted root cause set Caus before maintenance. After performing maintenance actions such as lubrication or replacement of the reducer on-site, traditional methods often only consider the maintenance effective based on whether the alarm event data Almr disappears. However, this solution further compares the parameter set Pbef before maintenance with the parameter set Paft after maintenance. If the friction parameter Fric does not decrease significantly, but the hysteresis parameter Back still maintains a large deviation, the falsification result set Fals can indicate that the maintenance did not truly hit the fault source. Then, by updating the knowledge set Know, the original judgment basis and maintenance correspondence are corrected, and finally, the health conclusion set Heal gives a clear conclusion of "maintenance not hit" or "continue observation". Therefore, this solution can specifically solve the problems of "lack of post-maintenance verification, continuous retention of erroneous maintenance experience, recurrence of the same fault, and continuous accumulation of cycle time fluctuations and quality risks" in the existing technology. It improves the fault prediction and health management of industrial robots from a single judgment to a closed-loop process with post-maintenance verification and continuous correction capabilities. It can reduce mis-repairs and repeated downtime, reduce the increase in maintenance costs caused by incorrect component replacement or incorrect maintenance actions, and improve the accuracy of subsequent similar fault judgments and maintenance decisions.

[0084] Example 2

[0085] Please see Figure 2 Specifically: the data processing module includes a data access unit;

[0086] The data access unit reads joint current data Jcur, joint temperature data Jtmp, following error data Ferr, alarm event data Almr, and program segment number data Segm from the industrial robot controller, and reads maintenance action data Mact from the maintenance terminal.

[0087] Then, establish associations based on the same industrial robot, the same program execution process, and the same maintenance process, and collect the reading results in a unified manner to form the original association set Raws;

[0088] It should be noted that:

[0089] Read joint current data Jcur, joint temperature data Jtmp, follower error data Ferr, alarm event data Almr, and program segment number data Segm from the industrial robot controller;

[0090] Among them, the joint current data Jcur is used to reflect the state of the execution drive load;

[0091] Joint temperature data Jtmp is used to reflect changes in heat accumulation.

[0092] The follow-up error data (Ferr) is used to reflect the execution deviation status;

[0093] Alarm event data (Almr) is used to reflect abnormal triggering status;

[0094] The program segment number data Segm is used to reflect the program segment to which the current operating condition belongs;

[0095] Then, read the maintenance action data Mact from the maintenance terminal to establish a correspondence between the maintenance execution process and the operation process before and after maintenance;

[0096] After the reading is completed, the joint current data Jcur, joint temperature data Jtmp, following error data Ferr, alarm event data Almr, program segment number data Segm, and maintenance action data Mact are uniformly collected according to the same industrial robot, the same program execution process, and the same maintenance process. This ensures that subsequent processing no longer deals with scattered data fragments from different sources, but instead obtains a raw correlation set Raws with a correlation basis.

[0097] Therefore, the final result of the data access unit is to output the raw association set Raws, which provides a unified input basis for subsequent execution time alignment and working condition classification.

[0098] The data processing module also includes a time-series classification unit;

[0099] The time series classification unit divides the operating condition interval Cseg based on the original association set Raws and according to the change order of the program segment number data Segm on the time axis.

[0100] Align the joint current data Jcur, joint temperature data Jtmp, follow-up error data Ferr, and alarm event data Almr, which fall within the same operating condition range Cseg, to the corresponding program segment number data Segm.

[0101] Then, the aligned joint current data Jcur, joint temperature data Jtmp, follow error data Ferr, alarm event data Almr, and corresponding program segment number data Segm are written into the runtime dataset Datr.

[0102] The maintenance interval Mwin is determined by the occurrence time of the first maintenance action data Mact and the last maintenance action data Mact in the same maintenance process, and the maintenance action data Mact located in the maintenance interval Mwin and the corresponding program segment number data Segm are written into the maintenance record set Drec.

[0103] Then, based on the program segment number data Segm corresponding to each working condition interval Cseg and the position of each working condition interval Cseg relative to the maintenance interval Mwin, a working condition label set Tagn is generated.

[0104] It should be noted that:

[0105] The operating condition interval Cseg is divided as follows: the time interval between two consecutive changes in the program segment number data Segm is defined as an operating condition interval Cseg; if the continuously collected program segment number data Segm remains unchanged, the corresponding time interval is assigned to the same operating condition interval Cseg.

[0106] The time alignment process is as follows: the acquisition time of the program segment number data Segm is used as the alignment reference time, and the joint current data Jcur, joint temperature data Jtmp, and follower error data Ferr under the same reference time are directly mapped to the reference time; when the occurrence time of the alarm event data Almr does not coincide with the reference time, the alarm event data Almr is mapped to the nearest reference time after the occurrence time.

[0107] The maintenance interval Mwin is determined as follows: the time when the first maintenance action data Mact occurs in the same maintenance process is determined as the start time, and the time when the last maintenance action data Mact occurs in the same maintenance process is determined as the end time. The time interval between the two is the maintenance interval Mwin.

[0108] The working condition label set Tagn is generated as follows: for each working condition interval Cseg, the program segment number data Segm corresponding to the working condition interval Cseg is recorded, and the sequence number of the working condition interval Cseg in the same program execution process is recorded. Then, the program segment number data Segm and the sequence number are combined as a working condition label and written into the working condition label set Tagn.

[0109] It should also be noted that: first, the operating condition interval Cseg is delineated from the program segment number data Segm; then, the joint current data Jcur, joint temperature data Jtmp, follow-up error data Ferr, and alarm event data Almr are written into the running dataset Datr according to the operating condition interval Cseg; then, the maintenance interval Mwin is determined from the maintenance action data Mact and written into the maintenance record set Drec; finally, the operating condition label set Tagn is generated based on the operating condition interval Cseg, the program segment number data Segm, and the sequence number of the operating condition interval Cseg in the same program execution process.

[0110] In this embodiment, through the data access unit and time sequence classification unit in the aforementioned data processing module, the joint current data Jcur, joint temperature data Jtmp, following error data Ferr, alarm event data Almr, program segment number data Segm, and maintenance action data Mact from the industrial robot controller and maintenance terminal can be uniformly collected into a raw association set Raws. Then, the operating condition interval Cseg is further divided, the maintenance interval Mwin is determined, and the running dataset Datr, maintenance record set Drec, and operating condition label set Tagn are formed. This allows the originally scattered, inconsistent, and different data to be organized into a unified data chain with the same program segment basis, the same time sequence basis, and the same maintenance process basis. The benefit of this is not to directly determine whether the maintenance hits the fault source, but to solidify the data foundation on which subsequent analysis depends, and to avoid mistaking data from different program segments, different time points, or even different maintenance processes as the same object for comparison. For example, in an automotive welding production line, the joint current data Jcur, joint temperature data Jtmp, and following error data Ferr of the same industrial robot in the spot welding program segment and the idle movement program segment are inherently significantly different. If the operating condition interval Cseg is not divided by the program segment number data Segm, the subsequently generated running dataset Datr will mix the two different operating states together. Similarly, if the maintenance interval Mwin is not determined by the first and last maintenance action data Mact, the data boundaries before and after maintenance will be confused, and the short-term recovery data after maintenance may even be mistakenly included in the analysis range before maintenance, leading to distortion in subsequent prediction modeling and parameter inversion.

[0111] Example 3

[0112] Please see Figure 3 Specifically: the predictive modeling module includes a baseline generation unit;

[0113] The baseline generation unit filters the labels whose corresponding working condition interval Cseg is located before the maintenance interval Mwin based on the working condition label set Tagn, and arranges the filtered labels in chronological order.

[0114] Then, remove a preset number of tags that are closest to the maintenance interval Mwin in terms of time sequence, and form the baseline tag set Btag with the remaining tags;

[0115] Extract the joint current data Jcur, joint temperature data Jtmp, follow error data Ferr, alarm event data Almr, and program segment number data Segm corresponding to the baseline tag set Btag from the running dataset Datr to form the baseline segment set Bref;

[0116] The data in the baseline segment set Bref that have the same working condition label set Tagn and the same program segment number data Segm are matched according to the time order. The average values ​​of the joint current data Jcur, joint temperature data Jtmp and follow-up error data Ferr at the corresponding positions are calculated to form the current reference set Curb, the temperature rise reference set Tmpb and the error reference set Errb.

[0117] Then, write the current reference set Curb, temperature rise reference set Tmpb, and error reference set Errb corresponding to each operating condition label set Tagn into the twin baseline set Bsim;

[0118] It should be noted that:

[0119] The baseline label set Btag is formed as follows: First, select the labels whose corresponding working condition interval Cseg is located before the maintenance interval Mwin from the working condition label set Tagn, and then arrange the selected labels in chronological order; then remove a preset number of labels that are closest to the maintenance interval Mwin in chronological order, and write the remaining labels into the baseline label set Btag.

[0120] The baseline fragment set Bref is formed by extracting the joint current data Jcur, joint temperature data Jtmp, follow error data Ferr, alarm event data Almr, and program segment number data Segm corresponding to the baseline tag set Btag from the running dataset Datr, and writing them into the baseline fragment set Bref in chronological order.

[0121] The current reference set Curb, the temperature rise reference set Tmpb, and the error reference set Errb are formed as follows:

[0122] For the data in the baseline fragment set Bref that correspond to the same label in the working condition label set Tagn and have the same program segment number data Segm, first arrange them in chronological order, and then use the data with the same sequential position as the same comparison position.

[0123] The average value of the joint current data Jcur at the same comparison position is used to obtain the current reference set Curb;

[0124] The average value of joint temperature data Jtmp at the same comparison position is used to obtain the temperature rise reference set Tmpb; the average value of following error data Ferr at the same comparison position is used to obtain the error reference set Errb.

[0125] The predictive modeling module also includes a root cause extraction unit;

[0126] The root cause extraction unit filters a preset number of tags whose corresponding working condition interval Cseg is located before the maintenance interval Mwin and is closest to the maintenance interval Mwin in time order, based on the working condition tag set Tagn and maintenance interval Mwin, to form a preceding tag set Ptag.

[0127] Extract the joint current data Jcur, joint temperature data Jtmp, follow-up error data Ferr, alarm event data Almr, and program segment number data Segm from the running dataset Datr, which correspond to the preceding tag set Ptag, to form the pre-maintenance segment set Sbef.

[0128] Extract the current reference set Curb, temperature rise reference set Tmpb, and error reference set Errb corresponding to the preceding tag set Ptag from the twin baseline set Bsim;

[0129] Subtract the joint current data Jcur in the pre-maintenance segment set Sbef from the corresponding current reference set Curb item by item according to the same program segment number data Segm and the same order position to form the current difference set Dcur.

[0130] Subtract the joint temperature data Jtmp in the pre-maintenance segment set Sbef from the corresponding temperature rise reference set Tmpb by subtracting the data with the same program segment number Segm and the same order of position to form the temperature rise difference set Dtmp.

[0131] Subtract the following error data Ferr in the pre-maintenance segment set Sbef from the corresponding error reference set Errb item by item according to the same program segment number data Segm and the same order position to form the error difference set Derr.

[0132] The current difference set Dcur, the temperature rise difference set Dtmp, and the error difference set Derr are combined according to the corresponding preceding label set Ptag, the corresponding program segment number data Segm, and the corresponding sequence position to form the deviation relationship set Devr;

[0133] Based on the deviation relationship set Devr and the corresponding alarm event data Almr, generate the hysteresis judgment value Bres, friction judgment value Fres, thermal drift judgment value Tres and delay judgment value Lres respectively.

[0134] Write the hysteresis judgment value Bress, friction judgment value Fress, thermal drift judgment value Tress, and delay judgment value Lres into the prediction root cause set Caus according to the corresponding preceding label set Ptag and the corresponding program segment number data Segm.

[0135] It should be noted that:

[0136] The preceding tag set Ptag is formed as follows: select the tags whose corresponding working condition interval Cseg is located before the maintenance interval Mwin from the working condition tag set Tagn, and then arrange the selected tags in chronological order; then write the preset number of tags that are closest to the maintenance interval Mwin in chronological order into the preceding tag set Ptag.

[0137] The preset number of tags is set as follows: based on the total number of available tags located before the maintenance interval Mwin and corresponding to the same program segment number Segm, preferably 1 to 5, or 10% to 30% of the total number of available tags rounded up; when the preset number of tags is used to form the baseline tag set Btag, preferably 2 to 5, in order to remove tags that are closest to the maintenance interval Mwin and are more likely to have entered the abnormal evolution stage; when the preset number of tags is used to form the preceding tag set Ptag, preferably 1 to 3, in order to retain a small number of tags closest to the maintenance interval Mwin as objects for pre-maintenance analysis; through the above setting method, it can be ensured that there are sufficient historical samples when constructing the baseline, and that the change characteristics of the period closest to the occurrence of the fault can be highlighted during the pre-maintenance analysis;

[0138] The pre-maintenance fragment set Sbef is formed by extracting the joint current data Jcur, joint temperature data Jtmp, follow-up error data Ferr, alarm event data Almr, and program segment number data Segm from the running dataset Datr, which correspond to the preceding tag set Ptag, and writing them into the pre-maintenance fragment set Sbef in chronological order.

[0139] The method for extracting the corresponding reference results from the twin baseline set Bsim is as follows: For each tag in the preceding tag set Ptag, read the current reference set Curb, temperature rise reference set Tmpb, and error reference set Errb corresponding to that tag from the twin baseline set Bsim, and retain the program segment number data Segm and sequence position corresponding to the read results.

[0140] The current difference set Dcur is formed by subtracting the joint current data Jcur in the pre-maintenance segment set Sbef from the corresponding current reference set Curb extracted from the twin baseline set Bsim, item by item, according to the same program segment number data Segm and the same order position, and writing the resulting difference into the current difference set Dcur.

[0141] The temperature rise difference set Dtmp is formed as follows: the joint temperature data Jtmp in the pre-maintenance segment set Sbef is subtracted from the corresponding temperature rise reference set Tmpb extracted from the twin baseline set Bsim, with the same program segment number data Segm and the same order position, and the resulting difference is written into the temperature rise difference set Dtmp.

[0142] The error difference set Derr is formed by subtracting the following error data Ferr in the pre-maintenance segment set Sbef from the corresponding error benchmark set Errb extracted from the twin baseline set Bsim, item by item, according to the same program segment number data Segm and the same order position, and writing the resulting difference into the error difference set Derr.

[0143] The deviation relation set Devr is formed by combining the current difference set Dcur, the temperature rise difference set Dtmp, and the error difference set Derr according to the corresponding preceding label set Ptag, the corresponding program segment number data Segm, and the corresponding sequence position, and writing the combination result into the deviation relation set Devr.

[0144] The hysteresis judgment value Bres is generated as follows: when the error difference set Delr corresponding to the deviation relation set Devr is greater than the preset error threshold at multiple consecutive sequential positions, and the current difference set Dcur corresponding to the same multiple consecutive sequential positions is not greater than the preset current threshold, the judgment result is written as the hysteresis judgment value Bres is valid; otherwise, the judgment result is written as the hysteresis judgment value Bres is invalid.

[0145] The preset error threshold is set as follows: Under the same operating condition label set Tagn and the same program segment number data Segm, extract the following error data Ferr or the error difference distribution obtained from it corresponding to the historical healthy period, and set the 90th to 95th percentile of the distribution as the preset error threshold, preferably the 95th percentile; when long-term historical data is lacking, 1.2 to 1.5 times the mean of the corresponding benchmark error can also be used as the preset error threshold; through the above setting method, the error fluctuations that are allowed under normal operating conditions can be distinguished from abnormal errors that are significantly beyond the healthy range, thereby avoiding misjudging a single small deviation as a fault symptom;

[0146] The friction judgment value Fres is generated as follows: when the current difference set Dcur and the temperature rise difference set Dtmp in the deviation relationship set Devr are continuously greater than their respective preset thresholds in the same order position, and the error difference set Derr in the same order position is also greater than the preset error threshold, the judgment result is written as the friction judgment value Fres is valid; otherwise, the judgment result is written as the friction judgment value Fres is invalid.

[0147] The preset thresholds are set as follows: for the threshold corresponding to the current difference set Dcur, the same setting method as the preset current threshold is used; for the threshold corresponding to the temperature rise difference set Dtmp, under the same operating condition label set Tagn and the same program segment number data Segm, the joint temperature data Jtmp corresponding to the historical healthy period or the temperature rise difference distribution obtained from it is extracted, and the 85th to 95th percentiles of the distribution are set as the corresponding thresholds; for the threshold corresponding to the error difference set Derr, the same setting method as the preset error threshold is used. Through the above setting method, different physical quantities can adopt judgment boundaries that are adapted to their own fluctuation characteristics, thereby avoiding the misjudgment problem caused by using the same fixed value to roughly compare the three types of data: current, temperature rise, and error.

[0148] The method for generating the thermal drift judgment value Tres is as follows: when the temperature rise difference set Dtmp corresponding to the deviation relation set Devr continues to increase in time sequence, and the error difference set Derr corresponding to the same time sequence increases synchronously, the judgment result is written as the thermal drift judgment value Tres is valid; otherwise, the judgment result is written as the thermal drift judgment value Tres is invalid.

[0149] The delay judgment value Lres is generated as follows: when the error difference set Derr in the deviation relation set Devr is greater than the preset error threshold in multiple consecutive sequential positions, and the corresponding alarm event data Almr indicates a following abnormality, the judgment result is written as the delay judgment value Lres is valid; otherwise, the judgment result is written as the delay judgment value Lres is invalid.

[0150] The prediction root cause set Caus is generated as follows: the hysteresis judgment value Bres, friction judgment value Fres, thermal drift judgment value Tres, and delay judgment value Lres are written into the prediction root cause set Caus according to the corresponding preceding tag set Ptag and the corresponding program segment number data Segm.

[0151] In this embodiment, through the baseline generation unit and root cause extraction unit in the aforementioned predictive modeling module, the baseline label set Btag can be first divided before the maintenance interval Mwin using the operating condition label set Tagn. This base tag set Btag is in the same operating condition as the current analysis object, but not directly close to the maintenance time. Based on this, the baseline fragment set Bref is extracted from the running dataset Datr to form the current reference set Curb, the temperature rise reference set Tmpb, and the error reference set Errb corresponding to each operating condition. Finally, these are written into the twin baseline set Bsim. Subsequently, the pre-maintenance fragment set Sbef is extracted from the preceding label set Ptag that is closest to the maintenance interval Mwin in time sequence, and the joint current data Jcur, joint temperature data Jtmp, and following data are extracted from it. The error data Ferr is compared item by item with the corresponding benchmark results in the twin baseline set Bsim to form the current difference set Dcur, the temperature rise difference set Dtmp, the error difference set Derr, and the deviation relationship set Devr. Then, the hysteresis judgment value Bres, the friction judgment value Fres, the thermal drift judgment value Tres, and the delay judgment value Lres are obtained and written into the predicted root cause set Caus. The advantage of this is that it does not simply judge the fault based on a single alarm or the magnitude of a single instantaneous parameter. Instead, it first establishes a "historical normal reference under the same operating conditions" and then quantitatively compares the abnormal changes before the maintenance, thereby distinguishing between "the program segment itself has a high load" and "the temperature rise will naturally increase" from "true abnormal deterioration". For example, in automotive welding scenarios, when the same industrial robot executes a welding program segment, the joint current data Jcur and joint temperature data Jtmp will naturally be higher than those in the idle movement program segment. Without a twin baseline set Bsim as a comparison under the same working conditions, it is easy to misjudge normal welding load as friction abnormality. However, with the above method, if it is found that the error difference set Derr increases continuously but the current difference set Dcur does not increase significantly at the same time, it is possible to more specifically determine that the hysteresis judgment value Bres is valid, rather than misjudging it as the friction judgment value Fres is valid. This makes the predicted root cause set Caus closer to the real degradation source, reducing incorrect maintenance, repeated downtime, and incorrect replacement of spare parts caused by root cause identification bias.

[0152] Example 4

[0153] Please see Figure 4 Specifically: the parameter inversion module includes a post-extraction unit;

[0154] The subsequent extraction unit, based on the working condition label set Tagn, the maintenance interval Mwin, and the preceding label set Ptag, filters labels whose corresponding working condition interval Cseg is located after the maintenance interval Mwin and whose working condition label is the same as the working condition label in the preceding label set Ptag, to form the subsequent label set Atag.

[0155] Then, extract the joint current data Jcur, joint temperature data Jtmp, follow-up error data Ferr, alarm event data Almr, and program segment number data Segm from the running dataset Datr to form the post-maintenance segment set Saft;

[0156] It should be noted that:

[0157] The subsequent label set Atag is formed as follows: Select the label whose corresponding working condition interval Cseg is located after the maintenance interval Mwin from the working condition label set Tagn, and then read the working condition label in the preceding label set Ptag; then, retain the label whose working condition label is the same as the working condition label in the preceding label set Ptag, and write the retained label into the subsequent label set Atag.

[0158] The purpose is to ensure that the data extraction range after maintenance is kept under the same program segment conditions as the analysis range before maintenance. By forming a subsequent label set Atag, it is possible to ensure that the subsequent post-maintenance fragment set Saft is comparable to the pre-maintenance fragment set Sbef, thereby improving the correspondence between the parameter inversion results before and after maintenance.

[0159] The post-maintenance fragment set Saft is formed as follows: extract the joint current data Jcur, joint temperature data Jtmp, follow-up error data Ferr, alarm event data Almr, and program segment number data Segm corresponding to the subsequent tag set Atag from the running dataset Datr, and write them into the post-maintenance fragment set Saft in chronological order.

[0160] The purpose is to organize the data that have been repaired and meet the same program segment conditions into a unified analysis object. By forming a post-repair segment set (Saft), a complete, continuous and consistent data foundation can be provided for subsequent post-repair parameter solving.

[0161] The parameter inversion module also includes a pre-parameter solving unit;

[0162] The pre-parameter solving unit extracts the current reference set Curb, temperature rise reference set Tmpb, and error reference set Errb corresponding to the pre-label set Ptag from the twin baseline set Bsim, forming the pre-reference segment set Bpre;

[0163] Subtract the joint current data Jcur in the pre-maintenance segment set Sbef from the current reference set Curb in the pre-maintenance reference segment set Bpre by subtracting each item from the data with the same program segment number Segm and the same order position to form the pre-current difference set Cpre.

[0164] Subtract the joint temperature data Jtmp in the pre-maintenance segment set Sbef from the temperature rise reference set Tmpb in the pre-maintenance reference segment set Bpre by subtracting each item from the data with the same program segment number Segm and the same order position to form the pre-temperature rise difference set Tpre.

[0165] Subtract the following error data Ferrr in the pre-maintenance segment set Sbef from the error reference set Errb in the pre-maintenance reference segment set Bpre by subtracting each item from the data with the same program segment number Segm and the same order position to form the pre-error difference set Epre.

[0166] Based on the alarm event data Almr corresponding to the previous current difference set Cpre, previous temperature rise difference set Tpre, previous error difference set Epre and previous maintenance segment set Sbef, the previous backlash value Bvbf, previous friction value Fvbf, previous thermal drift value Tvbf and previous delay value Lvbf are generated respectively.

[0167] Meanwhile, based on the previous backlash value Bvbf, previous friction value Fvbf, previous heat drift value Tvbf, and previous delay value Lvbf, the backlash parameter Back, friction parameter Fric, heat drift parameter Ther, and delay parameter Dely in the pre-maintenance parameter set Pbef are determined respectively, thus forming the pre-maintenance parameter set Pbef;

[0168] It should be noted that:

[0169] The pre-reference fragment set Bpre is formed as follows: For each tag in the preceding tag set Ptag, the corresponding current reference set Curb, temperature rise reference set Tmpb, and error reference set Errb are read from the twin baseline set Bsim. These references are then written into the pre-reference fragment set Bpre in tag order. The purpose is to establish a one-to-one correspondence between the pre-repair fragment set Sbef and the pre-reference fragment set Bpre. By forming the pre-reference fragment set Bpre, a clear correspondence can be ensured for subsequent difference calculations.

[0170] The pre-maintenance current difference set Cpre is formed as follows: the joint current data Jcur in the pre-maintenance segment set Sbef is subtracted item by item from the current reference set Curb in the pre-maintenance reference segment set Bpre, according to the same program segment number Segm and the same order of positions. The resulting difference is then written into the pre-maintenance current difference set Cpre. The purpose is to quantify the degree of deviation of the pre-maintenance current from the reference result. By forming the pre-maintenance current difference set Cpre, a direct basis can be provided for subsequent identification of friction changes.

[0171] The pre-maintenance temperature rise difference set Tpre is formed as follows: the joint temperature data Jtmp in the pre-maintenance segment set Sbef is subtracted item by item from the pre-maintenance reference segment set Bpre, using the same program segment number data Segm and the same order of positions. The resulting difference is then written into the pre-maintenance temperature rise difference set Tpre. The purpose is to quantify the degree of deviation of the pre-maintenance temperature rise from the reference result. By forming the pre-maintenance temperature rise difference set Tpre, a direct basis can be provided for subsequent identification of thermal drift changes and frictional heating changes.

[0172] The preceding error difference set Epre is formed as follows: the following error data Ferrr in the preceding segment set Sbef is subtracted item by item from the error benchmark set Errb in the preceding benchmark segment set Bpre, using the same program segment number data Segm and the same sequential position. The resulting difference is then written into the preceding error difference set Epre. The purpose of this process is to quantify the degree of deviation of the execution error before maintenance relative to the benchmark result. By forming the preceding error difference set Epre, a direct basis can be provided for subsequent identification of hysteresis changes and delay changes.

[0173] The preceding back-current difference (Bvbf) is generated as follows: The number of consecutive sequential positions in the preceding error difference set Epre that are greater than a preset error threshold, and the preceding current difference set Cpre corresponding to the same consecutive sequential positions, is not greater than a preset current threshold, is counted. The ratio of this count to the total number of sequential positions in the preceding error difference set Epre is then written as the preceding back-current difference (Bvbf). The purpose is to transform the characteristic of "error increasing but current not significantly increasing" into a quantifiable result. By generating the preceding back-current difference (Bvbf), it is possible to more clearly distinguish between back-current difference anomalies and load increase anomalies.

[0174] The preset current threshold is set as follows: under the same operating condition tag set Tagn and the same program segment number data Segm, extract the joint current data Jcur corresponding to the historical healthy period or the current difference distribution obtained from it, and set the 85th to 95th percentile of the distribution as the preset current threshold, preferably the 90th percentile; when long-term historical data is lacking, 1.1 to 1.4 times the corresponding reference current average can also be used as the preset current threshold; through the above setting method, normal load changes and abnormal load increases can be distinguished more accurately, avoiding misjudging the current increase caused by the process itself as abnormal friction or other faults;

[0175] The pre-friction value Fvbf is generated as follows: The number of times the pre-current difference set Cpre and the pre-temperature rise difference set Tpre are simultaneously greater than their respective preset thresholds at the same sequential position, and the number of times the pre-error difference set Epre corresponding to the same sequential position is simultaneously greater than the preset error threshold; then, the ratio of this number to the total number of sequential positions of the pre-current difference set Cpre is written as the pre-friction value Fvbf. The purpose is to transform the joint change characteristic of "increased current, increased temperature rise, and simultaneous increase in error" into a quantifiable result. By generating the pre-friction value Fvbf, the degree of friction abnormality before maintenance can be more clearly represented.

[0176] The preceding heat drift value Tvbf is generated as follows: The difference between two adjacent sequential positions in the preceding temperature rise difference set Tpre is calculated in chronological order, and the number of consecutive positive differences is counted. The ratio of the number of consecutive positive differences to the total number of adjacent differences is then written as the preceding heat drift value Tvbf. The purpose is to transform the trend of continuous temperature rise accumulation into a quantifiable result. By generating the preceding heat drift value Tvbf, the degree of accumulation of abnormal heat drift before maintenance can be more clearly represented.

[0177] The pre-delay value Lvbf is generated as follows: Multiple consecutive sequential positions in the pre-error difference set Epre are found to be greater than a preset error threshold, and the alarm event data Almr in the corresponding pre-maintenance segment set Sbef represents the number of times the anomaly occurred. The ratio of this number to the total number of sequential positions in the pre-error difference set Epre is then written as the pre-delay value Lvbf. The purpose is to transform the characteristic of "continuously increasing error accompanied by alarms following anomalies" into a quantifiable result. By generating the pre-delay value Lvbf, the degree of pre-maintenance delay anomaly can be more clearly represented.

[0178] The pre-maintenance parameter set Pbef is formed as follows: the backlash parameter Back is determined based on the previous backlash value Bvbf; the friction parameter Fric is determined based on the previous friction value Fvbf; the thermal drift parameter Ther is determined based on the previous thermal drift value Tvbf; and the delay parameter Dely is determined based on the previous delay value Lvbf. This achieves the unified conversion of various intermediate calculation results before maintenance into structured parameter results. By forming the pre-maintenance parameter set Pbef, direct input can be provided for subsequent consistency determination and falsification processing.

[0179] The parameter inversion module also includes a post-parameter solving unit;

[0180] The post-parameter solving unit extracts the current reference set Curb, the temperature rise reference set Tmpb, and the error reference set Errb corresponding to the subsequent tag set Atag from the twin baseline set Bsim, forming the post-reference segment set Baft;

[0181] Then, subtract the joint current data Jcur in the post-repair segment set Saft from the current reference set Curb in the post-repair reference segment set Baft item by item according to the same program segment number data Segm and the same order position to form the post-current difference set Caft.

[0182] Subtract the joint temperature data Jtmp in the post-maintenance segment set Saft from the temperature rise reference set Tmpb in the post-base segment set Baft, using the same program segment number data Segm and the same order position, to form the post-temperature rise difference set Taft.

[0183] Subtract the follow-up error data Ferrr in the post-maintenance segment set Saft from the error reference set Errb in the post-basic segment set Baft, item by item according to the same program segment number data Segm and the same order position, to form the post-error difference set Eaft;

[0184] Based on the alarm event data Almr corresponding to the post-current difference set Caft, post-temperature rise difference set Taft, post-error difference set Eaft, and post-maintenance segment set Saft, the post-hysteresis value Bvaf, post-friction value Fvaf, post-thermal drift value Tvaf, and post-delay value Lvaf are generated respectively.

[0185] At the same time, based on the back back value Bvaf, the back friction value Fvaf, the back heat drift value Tvaf, and the back delay value Lvaf, the back back parameter Back, the friction parameter Fric, the heat drift parameter Ther, and the delay parameter Dely in the post-maintenance parameter set Paft are determined respectively, thus forming the post-maintenance parameter set Paft;

[0186] It should be noted that:

[0187] The post-baseline fragment set Baft is formed as follows: For each tag in the subsequent tag set Atag, the current reference set Curb, temperature rise reference set Tmpb, and error reference set Errb corresponding to that tag are read from the twin baseline set Bsim, and the read current reference set Curb, temperature rise reference set Tmpb, and error reference set Errb are written into the post-baseline fragment set Baft in tag order; the purpose is to establish a one-to-one corresponding reference result for the post-maintenance fragment set Saft. By forming the post-baseline fragment set Baft, it can be ensured that the difference calculation after maintenance has a clear correspondence.

[0188] The post-maintenance current difference set Caft is formed as follows: the joint current data Jcur in the post-maintenance segment set Saft is subtracted from the current reference set Curb in the post-maintenance reference segment set Baft by subtracting each item from the data Segm with the same program segment number and the same order position, and the resulting difference is written into the post-maintenance current difference set Caft. The purpose is to quantify the degree of deviation of the post-maintenance current from the reference result. By forming the post-maintenance current difference set Caft, a basis can be provided for subsequent judgment on whether the friction change has fallen back after maintenance.

[0189] The post-maintenance temperature rise difference set Taft is formed as follows: the joint temperature data Jtmp in the post-maintenance segment set Saft is subtracted from the temperature rise reference set Tmpb in the post-maintenance reference segment set Baft by subtracting each item from the same program segment number data Segm and the same order position, and the resulting difference is written into the post-maintenance temperature rise difference set Taft. The purpose is to quantify the degree of deviation of the post-maintenance temperature rise from the reference result. By forming the post-maintenance temperature rise difference set Taft, a basis can be provided for subsequent judgment on whether the heat drift has fallen back after maintenance.

[0190] The post-error difference set Eaft is formed as follows: the following error data Ferr in the post-maintenance segment set Saft is subtracted from the error benchmark set Errb in the post-baseline segment set Baft by subtracting each item from the same program segment number data Segm and the same order position, and the resulting difference is written into the post-error difference set Eaft. The purpose is to quantify the degree of deviation of the post-maintenance execution error from the benchmark result. By forming the post-error difference set Eaft, a basis can be provided for subsequent judgment on whether the post-maintenance backlash and delay changes have fallen back.

[0191] The method for generating the back hysteresis value Bvaf is as follows: count the number of times that multiple consecutive sequential positions in the back error difference set Eaft are greater than a preset error threshold, and the number of times that the corresponding back current difference set Caft is not greater than a preset current threshold; then write the ratio of the number of times to the total number of sequential positions in the back error difference set Eaft as the back hysteresis value Bvaf; the purpose is to transform the degree of hysteresis anomaly after maintenance into a quantifiable result. By generating the back hysteresis value Bvaf, a basis can be provided for subsequent judgment on whether the hysteresis has improved after maintenance.

[0192] The post-friction value Fvaf is generated as follows: the number of times that the post-current difference set Catt and the post-temperature rise difference set Taft are simultaneously greater than their respective preset thresholds at the same sequential position, and the post-error difference set Eaft corresponding to the same sequential position is simultaneously greater than the preset error threshold; then the ratio of the number of times to the total number of sequential positions of the post-current difference set Catt is written as the post-friction value Fvaf; the purpose is to transform the degree of friction-related anomalies after maintenance into a quantifiable result. By generating the post-friction value Fvaf, a basis can be provided for subsequent judgment on whether friction has improved after maintenance.

[0193] The post-heat drift value Tvaf is generated as follows: the difference between two adjacent sequential positions of the post-temperature rise difference set Taft is calculated in chronological order, and the number of consecutive positive differences is counted; then the ratio of the number of consecutive positive differences to the total number of adjacent differences is written as the post-heat drift value Tvaf. The purpose is to transform the cumulative degree of heat drift after maintenance into a quantifiable result. By generating the post-heat drift value Tvaf, a basis can be provided for subsequent judgment on whether heat drift has been improved after maintenance.

[0194] The post-delay value Lvaf is generated as follows: Multiple consecutive sequential positions in the post-error difference set Eaft are greater than a preset error threshold, and the alarm event data Almr in the corresponding post-maintenance segment set Saft represents the number of times the anomaly occurred; then, the ratio of this number to the total number of sequential positions in the post-error difference set Eaft is written as the post-delay value Lvaf. The purpose is to transform the degree of post-maintenance delay anomalies into a quantifiable result. By generating the post-delay value Lvaf, a basis can be provided for subsequent judgment on whether the post-maintenance delay has improved.

[0195] The post-repair parameter set Paft is formed as follows: the backlash parameter Back is determined based on the backlash value Bvaf, the friction parameter Fric is determined based on the friction value Fvaf, the thermal drift parameter Ther is determined based on the thermal drift value Tvaf, and the delay parameter Dely is determined based on the delay value Lvaf. The purpose is to uniformly convert the various intermediate calculation results after repair into structured parameter results. By forming the post-repair parameter set Paft, direct input can be provided for subsequent consistency judgment and falsification write-back.

[0196] In this embodiment, through the post-extraction unit, pre-parameter solving unit, and post-parameter solving unit in the parameter inversion module, the post-repair fragment set Saft, which maintains the same program segment number data Segm condition as the pre-extraction tag set Ptag, can be extracted from the running dataset Datr using the post-extraction tag set Atag. Then, the pre-reference fragment set Bpre and the post-reference fragment set Baft are formed based on the twin baseline set Bsim, respectively. Furthermore, the pre-current difference set Cpre, the pre-temperature rise difference set Tpre, the pre-error difference set Epre, and the post-current difference set Caft, the post-temperature rise difference set Taft, and the post-error difference set Eaft are obtained. Finally, the pre-repair parameter set Pbef and the post-repair parameter set Paft are formed, respectively. This transforms the superficial phenomenon of "appearing abnormal before repair" and "appearing restored after repair" into directly comparable structured parameter results, such as the hysteresis parameter Back, the friction parameter Fric, the thermal drift parameter Ther, and the delay parameter D. The change in ely before and after maintenance; the benefits of this are different from those of the data processing and predictive modeling mentioned earlier. It does not emphasize data processing first, nor does it emphasize finding the root cause earlier. Instead, it emphasizes quantifying whether there is a real improvement before and after maintenance. This avoids misjudging the effectiveness of maintenance based solely on whether the alarm event data Almr disappears or whether the equipment temporarily resumes operation. For example, in an automotive welding production line, after maintenance personnel perform lubrication or component adjustment on an industrial robot, the alarm event data Almr may not be triggered for a short time. However, if the friction parameter Fric in the parameter set Paft after maintenance is still high and the thermal drift parameter Ther has not dropped, it means that the source of the fault has not been truly eliminated. Conversely, if the parameter set Paft after maintenance shows a significant decrease in the back hysteresis parameter Back, friction parameter Fric, thermal drift parameter Ther, and delay parameter Dely compared to the parameter set Pbef before maintenance, it can be proven that the maintenance has indeed improved the mechanical clearance, frictional heating, or execution lag issues.

[0197] Example 5

[0198] Please see Figure 5 Specifically: the falsification write-back module includes a consistency determination unit;

[0199] The consistency determination unit extracts the maintenance action data Mact corresponding to the current maintenance process from the maintenance record set Drec, and determines the maintenance object set Mobj based on the maintenance action data Mact;

[0200] Subtract the hysteresis parameter Back in the parameter set Pbef before maintenance from the hysteresis parameter Back in the parameter set Paft after maintenance to obtain the hysteresis change value Bchg;

[0201] The friction parameter Fric in the parameter set Pbef before maintenance is subtracted from the friction parameter Fric in the parameter set Paft after maintenance to form the friction change value Fchg;

[0202] Subtract the thermal drift parameter Ther in the parameter set Pbef before maintenance from the thermal drift parameter Ther in the parameter set Paft after maintenance to obtain the thermal drift change value Tchg;

[0203] Subtract the delay parameter Dely in the parameter set Pbef before maintenance from the delay parameter Dely in the parameter set Paft after maintenance to obtain the delay change value Lchg;

[0204] Write the hysteresis change value Bchg, friction change value Fchg, thermal drift change value Tchg, and delay change value Lchg into the parameter change set Vary;

[0205] Then, the predicted root cause set Caus, the maintenance object set Mobj, and the parameter change set Vary are compared item by item, and the consistent or inconsistent results obtained from each comparison are written into the consistency result set Cons.

[0206] Then, extract the inconsistent results from the consistent result set Cons and write the extracted inconsistent results into the falsification result set Fals;

[0207] It should be noted that:

[0208] The maintenance object set Mobj is formed as follows: the maintenance action data Mact corresponding to the current maintenance process is read from the maintenance record set Drec, the maintenance object name is extracted from the maintenance action data Mact, and the extracted maintenance object name is written into the maintenance object set Mobj. The purpose is to convert the maintenance actions in the maintenance records into directly comparable maintenance object results. By forming the maintenance object set Mobj, a basis can be provided for subsequent judgment on whether the predicted root cause is consistent with the actual maintenance object.

[0209] The hysteresis change value Bchg is formed by subtracting the hysteresis parameter Back from the hysteresis parameter Back in the parameter set Paft after maintenance from the hysteresis parameter Back in the parameter set Pbef before maintenance, and writing the result as the hysteresis change value Bchg. The purpose is to quantify the degree of hysteresis reduction before and after maintenance. By forming the hysteresis change value Bchg, a basis can be provided for subsequent judgment on whether hysteresis-type faults have been effectively handled.

[0210] The friction change value Fchg is formed by subtracting the friction parameter Fric in the parameter set Paft after maintenance from the friction parameter Fric in the parameter set Pbef before maintenance, and writing the result as the friction change value Fchg. The purpose is to quantify the degree of friction reduction before and after maintenance. By forming the friction change value Fchg, a basis can be provided for subsequent judgment on whether friction-related faults have been effectively handled.

[0211] The thermal drift change value Tchg is formed by subtracting the thermal drift parameter Ther in the parameter set Paft after maintenance from the thermal drift parameter Ther in the parameter set Pbef before maintenance, and writing the result as the thermal drift change value Tchg. The purpose is to quantify the degree of thermal drift reduction before and after maintenance. By forming the thermal drift change value Tchg, a basis can be provided for subsequent judgment on whether thermal drift faults have been effectively handled.

[0212] The delay change value Lchg is formed by subtracting the delay parameter Dely in the parameter set Paft after maintenance from the delay parameter Dely in the parameter set Pbef before maintenance, and writing the result as the delay change value Lchg. The purpose is to quantify the degree of delay reduction before and after maintenance. By forming the delay change value Lchg, a basis can be provided for subsequent judgment on whether delay-type faults have been effectively handled.

[0213] The parameter change set Vary is formed by combining the hysteresis change value Bchg, friction change value Fchg, thermal drift change value Tchg, and delay change value Lchg according to the same maintenance process, and writing the combination result into the parameter change set Vary. The purpose is to organize the results of multiple parameter changes into a unified comparison object. By forming the parameter change set Vary, a centralized input can be provided for subsequent item-by-item comparisons.

[0214] The consistency result set Cons is formed by matching each judgment result in the prediction root cause set Caus with each maintenance object in the maintenance object set Mobj, and then comparing the matching result with the corresponding parameter change result in the parameter change set Vary.

[0215] When the target parameter indicated by the root cause set Caus is consistent with the maintenance object in the maintenance object set Mobj, and the change value of the corresponding parameter in the parameter change set Vary is greater than the preset fallback threshold, the comparison result of this item is written as consistent.

[0216] The preset fallback threshold is set as follows: using the corresponding parameter in the parameter set Pbef before maintenance as a benchmark, it is set according to the fallback ratio after maintenance, preferably 10% to 50%; specifically, for the back hysteresis parameter Back and the delay parameter Dely, it is preferably 20% to 50%; for the friction parameter Fric, it is preferably 15% to 40%; and for the thermal drift parameter Ther, it is preferably 10% to 35%. When the decrease of the corresponding parameter after maintenance is greater than the preset fallback threshold, it indicates that the parameter has effectively fallen back after maintenance. Through the above setting method, the judgment of "whether maintenance has truly improved" can be transformed from a subjective judgment to a quantitative judgment, avoiding the misjudgment that maintenance has been effective based solely on the short-term disappearance of the alarm.

[0217] When the target parameter indicated by the predicted root cause set Caus is inconsistent with the maintenance object in the maintenance object set Mobj, or when the change value of the corresponding parameter in the parameter change set Vary is not greater than the preset fallback threshold, the comparison result is written as inconsistent. Then, all comparison results are written into the consistency result set Cons, thereby unifying the prediction result, maintenance object, and post-maintenance parameter fallback result into the same decision chain. By forming the consistency result set Cons, complete item-by-item comparison results can be obtained, instead of directly jumping to the falsification conclusion.

[0218] The falsification write-back module also includes a write-back output unit;

[0219] The write-back output unit reads the preset root cause correlation and preset maintenance strategy rules;

[0220] Then, based on the falsification result set Fals, the root cause association items corresponding to the inconsistent results are determined from the read preset root cause association relationships, and the determined root cause association items are corrected.

[0221] At the same time, based on the falsification result set Fals, the maintenance strategy items corresponding to the inconsistent results are determined from the read preset maintenance strategy rules, and the determined maintenance strategy items are corrected.

[0222] Then, write the revised root cause correlation items and revised maintenance strategy items into the updated knowledge set Know;

[0223] Then, based on the updated knowledge set Know and the post-repair parameter set Paft, the health conclusion set Heal is output;

[0224] It should be noted that:

[0225] The definition of the preset root cause correlation is: recording the preset correspondence between each predicted root cause and the corresponding parameter anomaly. The predicted root cause includes at least the back error corresponding to the back error parameter Back, the friction anomaly corresponding to the friction parameter Fric, the thermal drift anomaly corresponding to the thermal drift parameter Ther, and the delay anomaly corresponding to the delay parameter Dely.

[0226] The purpose of setting preset root cause correlations is to provide the original basis for the interpretation and subsequent falsification of the predicted root cause set Caus. By setting preset root cause correlations, the system can clearly identify which type of parameter anomaly corresponds to each type of predicted root cause, thus giving subsequent corrective actions a clear target.

[0227] The definition of the preset maintenance strategy rule is: to record the preset maintenance action relationship corresponding to each predicted root cause, wherein the preset maintenance action relationship includes at least the maintenance object and maintenance action corresponding to the predicted root cause;

[0228] The purpose of setting preset maintenance strategy rules is to provide the original basis for whether maintenance actions hit the predicted root cause. By setting preset maintenance strategy rules, the system can clearly identify which type of maintenance object and maintenance action should correspond to each type of predicted root cause, thereby giving subsequent corrective actions a clear direction.

[0229] The preset root cause correlation and preset maintenance strategy rules are read as follows: the root cause correlation content and maintenance strategy content are read from the rule content pre-stored in the falsification write-back module. The purpose is to ensure that subsequent correction actions are based on existing rules, rather than being generated directly from a blank state. By reading the preset root cause correlation and preset maintenance strategy rules, it can be ensured that the "correction" action has a clear object to be corrected.

[0230] The method for determining root cause associations based on the falsification result set Fals is as follows: read the predicted root cause items corresponding to inconsistent results from the falsification result set Fals, then search for the root cause association items corresponding to the predicted root cause items from the preset root cause association relationships, and determine the root cause association items found as root cause association items to be corrected. The purpose is to accurately map the falsification results to specific root cause association content. By determining the root cause association items to be corrected, it can be ensured that the subsequent correction actions are targeted.

[0231] The method for correcting root cause association terms is as follows: when the falsification result set Fals indicates that the predicted root cause is inconsistent with the direction of parameter change after repair, the applicability of the root cause association term to be corrected is reduced; when the falsification result set Fals indicates that the predicted root cause is inconsistent with the repair object, the original correspondence between the root cause association term to be corrected and the corresponding repair object is deleted. The purpose is to prevent the root cause correspondence that has been falsified from continuing to be used as the basis for subsequent judgments. By correcting the root cause association terms, the impact of erroneous root cause paths in subsequent predictions can be reduced.

[0232] The method for determining maintenance strategy items based on the falsification result set Fals is as follows: read the maintenance objects and maintenance actions corresponding to inconsistent results from the falsification result set Fals, then search for the maintenance strategy items corresponding to the maintenance objects and maintenance actions from the preset maintenance strategy rules, and determine the maintenance strategy items found as the maintenance strategy items to be corrected; the purpose is to accurately map the falsification results to specific maintenance strategy content, and by determining the maintenance strategy items to be corrected, it can be ensured that the subsequent correction actions are targeted.

[0233] The method for modifying maintenance strategy items is as follows: when the falsification result set Fals indicates that the maintenance object is inconsistent with the predicted root cause, the original correspondence between the maintenance strategy item to be modified and the predicted root cause is deleted; when the falsification result set Fals indicates that the maintenance object is consistent with the predicted root cause but the parameters after maintenance do not meet the preset fallback requirements, the applicability of the maintenance strategy item to be modified is reduced. The purpose is to prevent maintenance strategies that do not hit the fault source or have insufficient repair effect from continuing to be used as the basis for priority maintenance. By modifying maintenance strategy items, the degree of matching between subsequent maintenance actions and the actual fault source can be improved.

[0234] The updated knowledge set Know is formed by organizing the corrected root cause correlation items and corrected maintenance strategy items according to the same maintenance process and writing them into the updated knowledge set Know. The purpose is to save the falsified and corrected knowledge content as a result that can be called back. By forming the updated knowledge set Know, an updated basis can be provided for subsequent health judgments and the next maintenance decision.

[0235] The health conclusion set Heal is formed as follows: based on the corrected root cause correlation items and corrected maintenance strategy items in the updated knowledge set Know, the status of the back hysteresis parameter Back, friction parameter Fric, thermal drift parameter Ther, and delay parameter Dely in the post-maintenance parameter set Paft is determined; when all parameters are less than the corresponding preset health threshold, the determination result is written as health recovery.

[0236] The preset health threshold is set as follows: under the same operating condition label set Tagn and the same program segment number data Segm, the structured parameter distribution corresponding to the historical health period is extracted, and the 90th to 95th percentiles of the corresponding parameter distribution are set as the preset health threshold. When judging the back hysteresis parameter Back, friction parameter Fric, thermal drift parameter Ther, and delay parameter Dely respectively, the above method can be used to determine the corresponding preset health threshold. When long-term historical samples are lacking, 1.1 to 1.3 times the corresponding parameter baseline value can also be used as the preset health threshold. Through the above setting method, the output of the health conclusion set Heal can be based on the historical health interval, thereby more accurately distinguishing the three states of "health restored", "repair missed" and "continue observation".

[0237] When at least one parameter is greater than the corresponding preset health threshold and the corresponding item has been written into the falsification result set Fals, the judgment result will be written as maintenance missed.

[0238] When at least one parameter is greater than the corresponding preset health threshold and the corresponding item is not written into the falsification result set Fals, the judgment result will be written as continue observation;

[0239] Then, the results of each judgment are written into the health conclusion set Heal. The purpose is to convert the knowledge correction results and post-maintenance parameter results into executable health judgment results. By forming the health conclusion set Heal, direct conclusions can be provided for subsequent re-inspection, continued maintenance or resumption of operation.

[0240] In this embodiment, through the consistency judgment unit and write-back output unit in the above-mentioned falsification write-back module, the predicted root cause set Caus, the maintenance object set Mobj, and the parameter change set Vary obtained from the pre-maintenance parameter set Pbef and the post-maintenance parameter set Paft can be put into the same judgment link to generate the consistency result set Cons and the falsification result set Fals. Then, the read preset root cause association relationship and preset maintenance strategy rules are further modified in a targeted manner to form an updated knowledge set Know, and finally output the health conclusion set Heal. The benefits of this are different from those of the previous data processing, predictive modeling and parameter inversion. It is not just about making the data clearer, finding the root cause of the fault more accurately, or quantifying the parameter changes before and after maintenance. Instead, it makes the question of "whether this maintenance has truly hit the predicted root cause and whether the original judgment relationship and maintenance strategy should be continued to be used in the future when encountering similar situations" a closed-loop process of continuous correction. For example, in an automotive welding production line, an industrial robot is identified by the predicted root cause set Caus as having an abnormal friction parameter Fric. Lubrication is performed on the corresponding reducer according to existing preset maintenance strategy rules. However, if the friction change value Fchg is small after maintenance, and the thermal drift change value Tchg does not show a significant decrease, and the actual maintenance object is inconsistent with the predicted root cause set Caus, then the consistency result set Cons will give an inconsistent result. The falsification result set Fals will further mark this type of mismatch between "predicted root cause—maintenance object—parameter decrease result" separately. Subsequently, the write-back output unit will correct the corresponding preset root cause correlation and preset maintenance strategy rules to prevent the system from prioritizing the same symptoms as an abnormal friction parameter Fric and continuing to recommend the same maintenance actions. Conversely, if the hysteresis change value Bchg, friction change value Fchg, thermal drift change value Tchg, or delay change value Lchg after maintenance can form a consistent correspondence with the maintenance object, the system will retain the corresponding rules and give a more explicit health recovery judgment in the health conclusion set Heal. Therefore, this part can specifically solve the problems in the existing technology such as "the repair is completed but it is unknown whether the repair is correct, erroneous experience is continuously retained, similar faults are repeatedly misjudged, and the repair strategy becomes more and more biased with use". It enables the system to not only judge the current fault, but also continuously correct the subsequent root cause judgment basis and repair decision basis based on the actual results after the repair, which can reduce repeated disassembly and assembly, incorrect replacement of parts and continuous line stoppage.

[0241] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A fault prediction and health management system for industrial robots based on digital twins, characterized in that: It includes a data processing module, a predictive modeling module, a parameter inversion module, and a falsification and write-back module; The data processing module collects joint current data Jcur, joint temperature data Jtmp, following error data Ferr, alarm event data Almr, program segment number data Segm, and maintenance action data Mact from the industrial robot controller and maintenance terminal, and performs time alignment and working condition classification to form the running dataset Datr, the maintenance record set Drec, and the working condition label set Tagn. The predictive modeling module extracts baseline segments from the running dataset Datr based on the working condition label set Tagn and the maintenance interval Mwin, generating a twin baseline set Bsim; and extracts the pre-maintenance segment set Sbef from the running dataset Datr based on the working condition label set Tagn and the maintenance interval Mwin, and compares the pre-maintenance segment set Sbef with the twin baseline set Bsim to obtain the predicted root cause set Caus. The parameter inversion module performs parameter inversion on the pre-maintenance and post-maintenance operation segments based on the twin baseline set Bsim, obtaining the pre-maintenance parameter set Pbef and the post-maintenance parameter set Paft. The falsification write-back module reads the preset root cause correlation and preset maintenance strategy rules, and performs consistency judgment based on the predicted root cause set Caus, the pre-maintenance parameter set Pbef, the post-maintenance parameter set Paft, and the maintenance record set Drec to generate a falsification result set Fals; then, it modifies the read preset root cause correlation and preset maintenance strategy rules according to the falsification result set Fals to form an updated knowledge set Know. Then, based on the updated knowledge set Know and the post-repair parameter set Paft, the health conclusion set Heal is output.

2. The industrial robot fault prediction and health management system based on digital twin as described in claim 1, characterized in that: The data processing module includes a data access unit; The data access unit reads joint current data Jcur, joint temperature data Jtmp, following error data Ferr, alarm event data Almr, and program segment number data Segm from the industrial robot controller, and reads maintenance action data Mact from the maintenance terminal. Then, establish associations based on the same industrial robot, the same program execution process, and the same maintenance process, and collect the reading results in a unified manner to form the original association set Raws.

3. The industrial robot fault prediction and health management system based on digital twins according to claim 2, characterized in that: The data processing module also includes a time-series classification unit; The time series classification unit divides the operating condition interval Cseg based on the original association set Raws and according to the change order of the program segment number data Segm on the time axis. Align the joint current data Jcur, joint temperature data Jtmp, follow-up error data Ferr, and alarm event data Almr, which fall within the same operating condition range Cseg, to the corresponding program segment number data Segm. Then, the aligned joint current data Jcur, joint temperature data Jtmp, follow error data Ferr, alarm event data Almr, and corresponding program segment number data Segm are written into the runtime dataset Datr. The maintenance interval Mwin is determined by the occurrence time of the first maintenance action data Mact and the last maintenance action data Mact in the same maintenance process, and the maintenance action data Mact located in the maintenance interval Mwin and the corresponding program segment number data Segm are written into the maintenance record set Drec. Then, based on the program segment number data Segm corresponding to each working condition interval Cseg and the position of each working condition interval Cseg relative to the maintenance interval Mwin, a working condition label set Tagn is generated.

4. The industrial robot fault prediction and health management system based on digital twins according to claim 3, characterized in that: The predictive modeling module includes a baseline generation unit; The baseline generation unit filters the labels whose corresponding working condition interval Cseg is located before the maintenance interval Mwin based on the working condition label set Tagn, and arranges the filtered labels in chronological order. Then, remove a preset number of tags that are closest to the maintenance interval Mwin in terms of time sequence, and form the baseline tag set Btag with the remaining tags; Extract the joint current data Jcur, joint temperature data Jtmp, follow error data Ferr, alarm event data Almr, and program segment number data Segm corresponding to the baseline tag set Btag from the running dataset Datr to form the baseline segment set Bref; The data in the baseline segment set Bref that have the same working condition label set Tagn and the same program segment number data Segm are matched according to the time order. The average values ​​of the joint current data Jcur, joint temperature data Jtmp and follow-up error data Ferr at the corresponding positions are calculated to form the current reference set Curb, the temperature rise reference set Tmpb and the error reference set Errb. Then, write the current reference set Curb, temperature rise reference set Tmpb, and error reference set Errb corresponding to each operating condition label set Tagn into the twin baseline set Bsim.

5. The industrial robot fault prediction and health management system based on digital twin as described in claim 4, characterized in that: The predictive modeling module also includes a root cause extraction unit; The root cause extraction unit filters a preset number of tags whose corresponding working condition interval Cseg is located before the maintenance interval Mwin and is closest to the maintenance interval Mwin in time order, based on the working condition tag set Tagn and maintenance interval Mwin, to form a preceding tag set Ptag. Extract the joint current data Jcur, joint temperature data Jtmp, follow-up error data Ferr, alarm event data Almr, and program segment number data Segm from the running dataset Datr, which correspond to the preceding tag set Ptag, to form the pre-maintenance segment set Sbef. Extract the current reference set Curb, temperature rise reference set Tmpb, and error reference set Errb corresponding to the preceding tag set Ptag from the twin baseline set Bsim; Subtract the joint current data Jcur in the pre-maintenance segment set Sbef from the corresponding current reference set Curb item by item according to the same program segment number data Segm and the same order position to form the current difference set Dcur. Subtract the joint temperature data Jtmp in the pre-maintenance segment set Sbef from the corresponding temperature rise reference set Tmpb by subtracting the data with the same program segment number Segm and the same order of position to form the temperature rise difference set Dtmp. Subtract the following error data Ferr in the pre-maintenance segment set Sbef from the corresponding error reference set Errb item by item according to the same program segment number data Segm and the same order position to form the error difference set Derr. The current difference set Dcur, the temperature rise difference set Dtmp, and the error difference set Derr are combined according to the corresponding preceding label set Ptag, the corresponding program segment number data Segm, and the corresponding sequence position to form the deviation relationship set Devr; Based on the deviation relationship set Devr and the corresponding alarm event data Almr, generate the hysteresis judgment value Bres, friction judgment value Fres, thermal drift judgment value Tres and delay judgment value Lres respectively. Write the hysteresis judgment value Bres, friction judgment value Fres, thermal drift judgment value Tres, and delay judgment value Lres into the prediction root cause set Caus according to the corresponding preceding label set Ptag and the corresponding program segment number data Segm.

6. The industrial robot fault prediction and health management system based on digital twin as described in claim 5, characterized in that: The parameter inversion module includes a post-extraction unit; The subsequent extraction unit, based on the working condition label set Tagn, the maintenance interval Mwin, and the preceding label set Ptag, filters labels whose corresponding working condition interval Cseg is located after the maintenance interval Mwin and whose working condition label is the same as the working condition label in the preceding label set Ptag, to form the subsequent label set Atag. Then, extract the joint current data Jcur, joint temperature data Jtmp, follow-up error data Ferr, alarm event data Almr, and program segment number data Segm from the running dataset Datr to form the post-maintenance segment set Saft.

7. The industrial robot fault prediction and health management system based on digital twin as described in claim 6, characterized in that: The parameter inversion module also includes a pre-parameter solving unit; The pre-parameter solving unit extracts the current reference set Curb, temperature rise reference set Tmpb, and error reference set Errb corresponding to the pre-label set Ptag from the twin baseline set Bsim, forming the pre-reference segment set Bpre; Subtract the joint current data Jcur in the pre-maintenance segment set Sbef from the current reference set Curb in the pre-maintenance reference segment set Bpre by subtracting each item from the data with the same program segment number Segm and the same order position to form the pre-current difference set Cpre. Subtract the joint temperature data Jtmp in the pre-maintenance segment set Sbef from the temperature rise reference set Tmpb in the pre-maintenance reference segment set Bpre by subtracting each item from the data with the same program segment number Segm and the same order position to form the pre-temperature rise difference set Tpre. Subtract the following error data Ferrr in the pre-maintenance segment set Sbef from the error reference set Errb in the pre-maintenance reference segment set Bpre by subtracting each item from the data with the same program segment number Segm and the same order position to form the pre-error difference set Epre. Based on the alarm event data Almr corresponding to the previous current difference set Cpre, previous temperature rise difference set Tpre, previous error difference set Epre and previous maintenance segment set Sbef, the previous backlash value Bvbf, previous friction value Fvbf, previous thermal drift value Tvbf and previous delay value Lvbf are generated respectively. Simultaneously, based on the previous backlash value Bvbf, previous friction value Fvbf, previous thermal drift value Tvbf, and previous delay value Lvbf, the backlash parameter Back, friction parameter Fric, thermal drift parameter Ther, and delay parameter Dely in the pre-maintenance parameter set Pbef are determined respectively, thus forming the pre-maintenance parameter set Pbef.

8. The industrial robot fault prediction and health management system based on digital twin according to claim 7, characterized in that: The parameter inversion module also includes a post-parameter solving unit; The post-parameter solving unit extracts the current reference set Curb, the temperature rise reference set Tmpb, and the error reference set Errb corresponding to the subsequent tag set Atag from the twin baseline set Bsim, forming the post-reference segment set Baft; Then, subtract the joint current data Jcur in the post-repair segment set Saft from the current reference set Curb in the post-repair reference segment set Baft item by item according to the same program segment number data Segm and the same order position to form the post-current difference set Caft. Subtract the joint temperature data Jtmp in the post-maintenance segment set Saft from the temperature rise reference set Tmpb in the post-base segment set Baft, using the same program segment number data Segm and the same order position, to form the post-temperature rise difference set Taft. Subtract the follow-up error data Ferrr in the post-maintenance segment set Saft from the error reference set Errb in the post-basic segment set Baft, item by item according to the same program segment number data Segm and the same order position, to form the post-error difference set Eaft; Based on the alarm event data Almr corresponding to the post-current difference set Caft, post-temperature rise difference set Taft, post-error difference set Eaft, and post-maintenance segment set Saft, the post-hysteresis value Bvaf, post-friction value Fvaf, post-thermal drift value Tvaf, and post-delay value Lvaf are generated respectively. Simultaneously, based on the back back value Bvaf, back friction value Fvaf, back thermal drift value Tvaf, and back delay value Lvaf, the back back parameter Back, friction parameter Fric, thermal drift parameter Ther, and delay parameter Dely in the post-maintenance parameter set Paft are determined respectively, thus forming the post-maintenance parameter set Paft.

9. The industrial robot fault prediction and health management system based on digital twin as described in claim 8, characterized in that: The falsification write-back module includes a consistency determination unit; The consistency determination unit extracts the maintenance action data Mact corresponding to the current maintenance process from the maintenance record set Drec, and determines the maintenance object set Mobj based on the maintenance action data Mact; Subtract the hysteresis parameter Back in the parameter set Pbef before maintenance from the hysteresis parameter Back in the parameter set Paft after maintenance to obtain the hysteresis change value Bchg; The friction parameter Fric in the parameter set Pbef before maintenance is subtracted from the friction parameter Fric in the parameter set Paft after maintenance to form the friction change value Fchg; Subtract the thermal drift parameter Ther in the parameter set Pbef before maintenance from the thermal drift parameter Ther in the parameter set Paft after maintenance to obtain the thermal drift change value Tchg; Subtract the delay parameter Dely in the parameter set Pbef before maintenance from the delay parameter Dely in the parameter set Paft after maintenance to obtain the delay change value Lchg; Write the hysteresis change value Bchg, friction change value Fchg, thermal drift change value Tchg, and delay change value Lchg into the parameter change set Vary; Then, the predicted root cause set Caus, the maintenance object set Mobj, and the parameter change set Vary are compared item by item, and the consistent or inconsistent results obtained from each comparison are written into the consistency result set Cons. Then, extract the inconsistent results from the consistent result set Cons and write the extracted inconsistent results into the falsification result set Fals.

10. The industrial robot fault prediction and health management system based on digital twin as described in claim 9, characterized in that: The falsification write-back module also includes a write-back output unit; The write-back output unit reads the preset root cause correlation and preset maintenance strategy rules; Then, based on the falsification result set Fals, the root cause association items corresponding to the inconsistent results are determined from the read preset root cause association relationships, and the determined root cause association items are corrected. At the same time, based on the falsification result set Fals, the maintenance strategy items corresponding to the inconsistent results are determined from the read preset maintenance strategy rules, and the determined maintenance strategy items are corrected. Then, write the revised root cause correlation items and revised maintenance strategy items into the updated knowledge set Know; Then, based on the updated knowledge set Know and the post-repair parameter set Paft, the health conclusion set Heal is output.

Images