An intelligent manufacturing production line equipment fault online diagnosis method

Abstract

The application provides a kind of intelligent manufacturing production line equipment fault online diagnosis method, comprising: according to the diagnosis sequence, extract impact energy attenuation form and engagement sideband broadening characteristics, obtain damage symptom vector;According to the damage symptom vector, the components of the transmission chain are identified step by step, the current stage of wear degradation is identified, and the spatial coordinates of the fault source in the transmission chain are determined;When the evaluated fault influence domain has covered the key section, the wear mechanism type of the fault component is identified, the remaining operation life grade of the fault component is determined;After analyzing that the remaining operation life grade has fallen below the safety threshold, the optimal shutdown window under the current injection molding process beat is determined, and the replacement instruction with maintenance timing constraints is generated.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to an online fault diagnosis method for intelligent manufacturing production line equipment. Background Technology

[0002] The transmission system of injection molding equipment transmits driving force to the screw and mold clamping mechanism step by step through multi-stage gear meshing and bearing support. Each gear stage experiences cyclic contact stress during continuous high-load meshing, while the bearing raceways bear combined radial and axial loads during high-speed rotation. Each friction pair generates metal abrasive particles during continuous operation, which then flow into the lubricating oil circulation loop. The operating status of each component in the transmission system directly determines the mold clamping accuracy and plasticization uniformity of the injection molding process. Wear and degradation of any transmission component will lead to a decrease in the dimensional consistency and surface quality of the final product. Existing transmission system condition monitoring methods mostly rely on periodic shutdown inspections and fixed-cycle replacement strategies, lacking the ability to continuously track the gradual process of wear and degradation within the transmission chain. When the wear of a bearing or gear accelerates between two timed inspections, the fixed-time-cycle strategy struggles to capture the true process of wear degradation from its early onset to its accelerated expansion. This often results in premature replacement of components with remaining lifespan, leading to waste, or continued operation according to the original time cycle even when components are nearing failure. There is a significant discrepancy between maintenance timing and the actual wear state of the transmission components. Furthermore, the transmission chain is a multi-stage series structure, and the vibration signal obtained from a single monitoring location contains superimposed response components from each stage of the component. When early localized wear occurs in a component of a certain stage, its weak fault cycle response is masked by the speed frequency components generated by the normal operation of adjacent stages. The vibration amplitude from a single external detection point alone cannot determine which stage of the transmission chain the fault source is located in, or what stage the wear has progressed to. Localized wear develops from initial pitting to spalling and expansion. The temperature rise at the faulty component and the rate of abrasive particle formation intensify simultaneously. Heat is conducted along the transmission axis to adjacent load-bearing structures, and abrasive particles diffuse to other friction pairs with the lubricating oil circulation. Because maintenance decisions for transmission components are made solely based on a single vibration amplitude and a fixed replacement cycle, it is impossible to determine whether the fault thermal domain has spread to critical sections of the load-bearing structure, whether the wear mechanism has changed from normal break-in to abnormal wear mode, and the determination of the maintenance window lacks coordination with the current injection molding process cycle. This may lead to forced shutdowns midway through the injection molding cycle, resulting in the scrapping of work-in-process products. Summary of the Invention

[0003] This invention provides an online fault diagnosis method for intelligent manufacturing production line equipment, mainly including: The vibration envelope of the bearing raceway and the acoustic emission pulse of gear meshing of the rotating mechanism are obtained. The speed and power frequency interference data in the vibration envelope of the bearing raceway are stripped away. The fault cycle response embedded in the acoustic emission pulse of gear meshing is identified to obtain a diagnostic sequence. Based on the diagnostic sequence, the impact energy attenuation morphology and meshing sideband widening characteristics are extracted to obtain the damage sign vector; Based on the damage symptom vector, each level of the transmission chain components are identified step by step to determine the current stage of wear and degradation and to determine the spatial coordinates of the fault source in the transmission chain. Based on the spatial coordinates, locate the installation position of the faulty component, obtain the thermal infrared temperature field distribution and lubricating oil metal abrasive particle spectrum at the installation position, and assess whether the fault influence domain has covered the critical cross section of the injection molding equipment's load-bearing structure. When the assessment shows that the fault impact area has covered the critical section, the wear mechanism type of the faulty component is identified, and the remaining service life level of the faulty component is determined. Analysis shows that the remaining service life level has dropped below the safety threshold. The optimal shutdown window under the current injection molding cycle time is determined, and a part replacement instruction is generated. According to the replacement instruction, the faulty component is replaced. After replacement, assembly, calibration, and restart, the bearing raceway vibration envelope pattern reaches the steady-state operation stage. When it is confirmed that the vibration envelope pattern has returned to the preset healthy operating range, the injection molding equipment is restored to normal production cycle.

[0004] Furthermore, the process of acquiring the bearing raceway vibration envelope and gear meshing acoustic emission pulses of the rotating mechanism, stripping the speed and power frequency interference data from the bearing raceway vibration envelope, identifying the fault cycle response embedded in the gear meshing acoustic emission pulses, and obtaining a diagnostic sequence includes: The original vibration signal is demodulated using Hilbert transform to obtain the bearing raceway vibration envelope. The elastic stress wave is bandpass filtered to obtain the gear meshing acoustic emission pulse. The bearing raceway vibration envelope is subjected to adaptive mode decomposition to remove the speed and power frequency interference components, and retain the intrinsic components of the bearing raceway fault characteristic frequencies to obtain the raceway cleanup envelope. The abnormal transient impact response and the fault period of the raceway cleanup envelope are aligned and fused along the same time axis to form a diagnostic sequence.

[0005] Furthermore, the step of extracting the impact energy attenuation morphology and meshing sideband widening features based on the diagnostic sequence to obtain a damage sign vector includes: The amplitude decay trajectory of the vibration-side impact waveform in the diagnostic sequence is fitted into a one-sided exponential decay curve to obtain the decay time constant and the initial impact amplitude, thus forming the impact energy decay morphology characteristics. The sideband spectral clusters on both sides of the fundamental frequency of the acoustic emission side meshing pulse are located by short-time Fourier transform to obtain the meshing sideband broadening characteristics. The impact energy decay morphology characteristics are assigned to the bearing channel and the meshing sideband broadening characteristics are assigned to the gear channel. After normalization, they are concatenated into a multidimensional column vector to obtain the damage sign vector.

[0006] Furthermore, the step of identifying each stage of the transmission chain components according to the damage symptom vector, recognizing the current stage of wear and degradation, and determining the spatial coordinates of the fault source in the transmission chain includes: The bearing channel component and gear channel component in the damage symptom vector are filtered by the rotation frequency and meshing frequency window of each level of component to obtain the symptom response slices of each level of component; the symptom response slices are input into a support vector machine classifier trained with four types of working condition samples: normal running-in, early pitting, moderate spalling and severe failure, and the working condition category label of each level of component is output; the level number that deviates from the normal running-in category is mapped to the shaft segment number and bearing seat number on the transmission chain assembly drawing.

[0007] Furthermore, the step of locating the installation position of the faulty component based on the spatial coordinates, obtaining the thermal infrared temperature field distribution and lubricating oil metal abrasive particle spectrum at the installation position, and assessing whether the fault influence domain has covered the critical cross-section of the injection molding equipment's load-bearing structure includes: Based on the shaft segment number and bearing seat number, the corresponding physical installation position is retrieved from the digital assembly map of the injection molding equipment as the on-site calibration coordinates. Within the neighborhood of the on-site calibration coordinates, a thermal infrared imager is used to collect the radiation infrared intensity matrix, which is then converted into a temperature field matrix through blackbody radiation calibration. Simultaneously, oil samples are extracted from the lubricating oil circulation loop to measure the emission spectral intensity of iron, copper, and chromium elements, obtaining the metal abrasive spectra of the lubricating oil. The temperature field matrix is ​​used to determine the abnormal temperature rise region using the Otsu threshold method, and the isothermal closed contour of the abnormal temperature rise region is extracted as the temperature field diffusion boundary morphology. The temperature field diffusion boundary morphology is geometrically compared with the spatial projection areas of the tie rod root section, the elbow hinge pin section, and the frame load-bearing beam section to determine whether the fault influence domain has covered the critical sections of the injection molding equipment's load-bearing structure.

[0008] Furthermore, determining whether the fault's influence area covers the critical cross-section of the injection molding equipment's load-bearing structure includes: Steady-state temperature rise values ​​at thermal equilibrium are read from the normal break-in condition records for the tie rod root section, the elbow hinge pin section, and the frame load-bearing beam section. The average value during the break-in period is taken as the baseline amplitude of the temperature rise. The measured temperature rise amplitude values ​​of each section are read from the temperature field matrix of the current fault condition. The temperature rise offset is obtained by subtracting the corresponding baseline amplitude value from the measured temperature rise amplitude value of each section. When the temperature rise offset falls into the shallow penetration range, it is determined that the current wear degradation belongs to the normal break-in stage. When it falls into the medium penetration range or the deep penetration range, it is determined that it belongs to the abnormal failure stage.

[0009] Furthermore, when the assessed fault impact domain has covered the critical cross-section, the wear mechanism type of the faulty component is identified, and the remaining service life level of the faulty component is determined, including: The extracted oil sample is deposited under a magnetic field using a ferrometer and then scanned and imaged using a dual-light source optical microscope to obtain a set of ferrographic particle microscopic images. For each particle in the ferrographic particle microscopic image set, four morphological descriptive parameters are extracted: aspect ratio, edge sharpness, surface texture roughness, and thickness shadow distribution. These parameters are then input into a convolutional neural network classifier trained with three types of samples: cutting wear, fatigue spalling, and adhesive wear, to obtain the wear mechanism type. The remaining service life level of the faulty component is determined by querying a life level mapping table based on the wear mechanism type and the penetration depth level.

[0010] Furthermore, the step of querying the life grade mapping table by the wear mechanism type and penetration depth level to determine the remaining operating life level of the faulty component includes: the life grade mapping table uses cutting wear, fatigue spalling, and adhesive wear as row indexes, and shallow penetration level, medium penetration level, and deep penetration level as column indexes, and the table is filled with three levels of remaining operating life level: sufficient margin, tight margin, and near failure; the corresponding remaining operating life level is read from the intersection of the wear mechanism type and penetration depth level in the life grade mapping table.

[0011] Furthermore, after analysis, the remaining service life level has dropped below the safety threshold. The optimal shutdown window under the current injection molding cycle time is determined, and a part change instruction is generated, including: The remaining service life level is compared with the safety threshold set as the margin tight level. When the margin tightness level or near failure level is reached, the shutdown window selection process is triggered. The injection molding process cycle sequence containing the start and end timestamps of each stage of mold closing, injection holding, cooling, mold opening, and ejection is collected from the programmable logic controller of the injection molding equipment. The idle time segment from the end of cooling between each injection cycle to the start of the next mold closing is scanned. The longest time segment that satisfies the shortest replacement time for the faulty part is selected as the optimal shutdown window. Based on the start and end times of the optimal shutdown window and the spatial coordinates of the faulty part, a replacement instruction message containing shutdown time field, part location field, and replacement process sequence field is compiled and sent through the fieldbus to obtain the replacement instruction.

[0012] Furthermore, the step of performing the faulty component replacement operation according to the replacement instruction, extracting the bearing raceway vibration envelope pattern after replacement, assembly, calibration, and restart to reach a steady-state operation stage, and restoring the injection molding equipment to normal production cycle when it is confirmed that the vibration envelope pattern has returned to the preset healthy operating range, includes: The injection molding equipment is stopped and the assembly robot arm performs the removal of faulty parts, hoisting of new parts, and centering and calibration to obtain a calibrated transmission system. When the transmission system is restarted until it enters a steady-state operation stage, the vibration signal of the bearing housing is collected and demodulated to extract the vibration envelope of the bearing raceway under the operation of the new part. The upper and lower limits of the amplitude envelope during the break-in period are retrieved as a preset healthy operating range. When the vibration envelope of the bearing raceway falls within the preset healthy operating range throughout the process, the fault is determined to be eliminated, and the injection molding equipment is restored to normal production rhythm.

[0013] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses an online fault diagnosis method for intelligent manufacturing production line equipment. Addressing the challenges of early identification and accurate assessment of the impact domain of latent faults caused by wear and degradation in injection molding equipment transmission systems during long-term operation, the method collects bearing raceway vibration envelopes and gear meshing acoustic emission pulses. Adaptive modal decomposition is used to remove speed-frequency interference, and a damage symptom vector is constructed by fusing impact energy attenuation patterns and meshing sideband widening characteristics. This vector is then used to identify the spatial coordinates of the fault source at each stage of the transmission chain. Furthermore, by combining the thermal infrared temperature field diffusion boundary and lubricating oil ferrographic particle morphology at the installation location, the wear mechanism type and penetration depth level are determined. The method assesses whether key sections such as the tie rod root, crank pin, and frame beam are covered by the fault impact domain, thereby determining the remaining lifespan level and generating a replacement instruction with time constraints. After replacement, the vibration envelope returning to the healthy range is used as a closed-loop criterion. This achieves accurate end-to-end diagnosis from symptom detection to shutdown decision-making, significantly improving production line reliability and maintenance economy. Attached Figure Description

[0014] Figure 1 This is a flowchart of an online fault diagnosis method for intelligent manufacturing production line equipment according to the present invention.

[0015] Figure 2 This is a schematic diagram of an online fault diagnosis method for intelligent manufacturing production line equipment according to the present invention.

[0016] Figure 3 This is another schematic diagram of an online fault diagnosis method for intelligent manufacturing production line equipment according to the present invention. Detailed Implementation

[0017] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0018] like Figures 1-3 This embodiment of an online fault diagnosis method for intelligent manufacturing production line equipment may specifically include: S101. Obtain the bearing raceway vibration envelope and gear meshing acoustic emission pulse of the rotating mechanism in the transmission system of the injection molding equipment, strip the speed power frequency interference part of the bearing raceway vibration envelope, identify the fault cycle response embedded in the gear meshing acoustic emission pulse, and obtain the diagnostic sequence.

[0019] An accelerometer is installed at the shaft end of the rotating mechanism in the injection molding equipment transmission system to collect the original vibration signal of the bearing raceway under combined radial and axial loads. The original vibration signal is demodulated using Hilbert transform, and the amplitude modulation component of the high-frequency band is extracted to obtain the bearing raceway vibration envelope. An acoustic emission sensor is attached to the outer wall of the gearbox to capture the elastic stress wave released in the tooth contact area during gear meshing. After bandpass filtering to remove environmental noise, the gear meshing acoustic emission pulse is obtained. For the superimposed speed frequency component and the meshing frequency component of adjacent transmission stages in the bearing raceway vibration envelope, an adaptive modal decomposition method is used to decompose the vibration envelope into several intrinsic modal components. The speed frequency interference component is eliminated according to the frequency band overlap between each intrinsic modal component and the speed frequency, and the intrinsic component bearing the characteristic frequency of raceway faults is retained to obtain the raceway cleanup envelope. From the gear meshing acoustic emission pulses, pulse segments are extracted according to an integer multiple window of the gear meshing period. Abnormal transient impact responses embedded between the intrinsic meshing pulses are identified. The time coordinates of the abnormal transient impact responses are aligned and merged with the fault cycle time in the raceway purification envelope along the same time axis to form a diagnostic sequence.

[0020] In the transmission system of injection molding equipment, the rotating mechanism includes the output shaft of the drive motor, the first-stage reduction gear shaft, the second-stage intermediate shaft, and the power output shaft that finally connects to the screw or mold clamping screw. Both ends of each shaft segment are supported by deep groove ball bearings or tapered roller bearings.

[0021] For example, the accelerometer employs a piezoelectric structure and is attached to the radial bearing direction of the bearing housing via a magnetic base or threaded base. The sampling rate covers more than ten times the gear meshing frequency, capturing the impact vibration generated by the bearing raceway when the rolling element passes through the defect point.

[0022] It should be noted that the early pitting fault signal of bearing raceway exhibits high-frequency modulation characteristics, that is, the impact energy modulates the amplitude of high-frequency natural vibration with the bearing outer ring passing frequency, inner ring passing frequency or rolling element rotation frequency.

[0023] For example, the original vibration signal is demodulated by Hilbert transform, the signal is regarded as a real component and its imaginary orthogonal component is constructed, and the modulus value is taken to obtain the instantaneous amplitude curve. This curve strips out the high frequency component of the carrier and retains the low frequency modulation information, forming the vibration envelope of the bearing raceway.

[0024] In one possible implementation, the acoustic emission sensor on the outer wall of the gearbox is a resonant broadband sensor, with the resonant center frequency within the dominant frequency range of gear meshing impact. After the sensor output signal is amplified by a preamplifier, it is filtered by a bandpass filter to remove low-frequency mechanical noise and high-frequency electromagnetic interference, retaining the transient elastic stress wave pulses released during tooth surface spalling and pitting initiation, thus obtaining the gear meshing acoustic emission pulse.

[0025] Specifically, the adaptive mode decomposition method processes the following steps: the bearing raceway vibration envelope is considered as the sum of several intrinsic mode components and residual terms; all local maxima and local minima of the signal are repeatedly identified; upper and lower envelopes are constructed using cubic spline interpolation; the arithmetic mean of the two is taken to obtain the instantaneous mean curve; the instantaneous mean curve is subtracted from the current signal to obtain candidate components; the above screening operation is repeated for the candidate components until the absolute value of the difference between the number of extreme points and the number of zero crossing points of the candidate component does not exceed 1, and the absolute amplitude of the upper and lower envelope means is less than 0.01 times the peak value of the original signal; it is then identified as an intrinsic mode component and removed from the original signal; the remaining part is further screened until the residual terms show a monotonic trend or the amplitude decays to a negligible level, thereby obtaining several intrinsic mode components arranged from high to low frequency.

[0026] Preferably, the process of eliminating the power frequency interference component of the driving motor speed is as follows: the power frequency fs corresponding to the driving motor speed can be directly read from the feedback value of the servo encoder; the Fourier transform of each intrinsic mode component is performed to obtain its main spectrum peak frequency fi; if the relative deviation between fi and fs or its 2nd and 3rd harmonic frequencies satisfies Where k takes the values ​​1, 2, or 3, the intrinsic modal component is determined to carry the power frequency interference component of the rotational speed and is removed. Among the remaining intrinsic modal components, those whose main spectral peak is located in the neighborhood of the bearing outer ring passing frequency or inner ring passing frequency are retained and superimposed to reconstruct the raceway purification envelope. Furthermore, there is a steady-state meshing intrinsic pulse in the gear meshing process, whose repetition period is equal to the reciprocal of the gear rotational frequency divided by the number of teeth.

[0027] For example, the gear meshing acoustic emission pulses are slidably intercepted with an integer multiple of the gear meshing cycle as the window width. The kurtosis index of the pulse peak sequence is calculated in each window. The average kurtosis of the normal meshing window recorded during the steady-state operation phase is used as the baseline value K0. If the current window kurtosis K satisfies K>3*K0, the pulse segment in the window is determined to be an abnormal transient impact response, and its corresponding timestamp is recorded.

[0028] In one embodiment, four times the root mean square value of the raceway purification envelope during steady-state operation is taken as the amplitude judgment threshold Ath. The timestamps corresponding to the sampling points in the purification envelope whose amplitude exceeds Ath are regarded as fault cycle moments and included in the fault moment set. The timestamp set of the abnormal transient impact response and the fault moment set are projected onto the time axis of the same sampling clock reference and arranged in ascending order of timestamps to form a dual-channel event stream containing vibration side feature markers and acoustic emission side feature markers, i.e., a diagnostic sequence.

[0029] It is understood that the diagnostic sequence simultaneously carries the time-domain impact information of bearing raceway fault response and the transient stress wave information of gear meshing abnormality. The relative positional relationship of the two types of responses on the same time axis reflects the coupling state of multi-stage components in the transmission chain, providing a data basis for subsequent damage sign feature extraction.

[0030] S102. Based on the diagnostic sequence, extract the impact energy attenuation morphology and meshing sideband widening characteristics to obtain the damage sign vector.

[0031] The impact time-domain waveform segments corresponding to the vibration-side feature markers in the diagnostic sequence are extracted. The amplitude decay trajectory following the impact point is acquired point-by-point along the time axis. The amplitude decay trajectory is fitted into a one-sided exponential decay curve using least-squares fitting. The decay time constant and initial impact amplitude of the exponential decay curve are read to constitute the impact energy decay morphology feature. The meshing pulse segments corresponding to the acoustic emission-side feature markers in the diagnostic sequence are extracted. A short-time Fourier transform is performed on the meshing pulse segments to locate symmetrically distributed sideband spectral line clusters on both sides of the preset gear meshing base spectral line in the spectrum. The frequency offset of the sideband spectral line clusters relative to the meshing base frequency and the ratio of the sideband peak amplitude to the meshing base spectral peak amplitude are measured to obtain the meshing sideband broadening feature. The impact energy decay morphology feature is assigned to the bearing channel, and the meshing sideband broadening feature is assigned to the gear channel. The values ​​in both channels are normalized to the same numerical range using range normalization. The values ​​are then concatenated and spliced ​​into a multi-dimensional column vector in the order of bearing channel first, gear channel second, to obtain the damage sign vector.

[0032] During the operation of the transmission system of injection molding equipment, when pitting defects occur in the bearing raceway, each time the rolling element passes through the defect point, it will generate a transient impact. The impact energy is then transmitted to the adjacent structure along the bearing path and gradually attenuates. When tooth surface spalling occurs during gear meshing, the spalling point modulates the phase and amplitude of the meshing vibration during meshing, which is manifested in the spectrum as a group of sideband spectral lines symmetrically distributed around the meshing fundamental frequency. The impact energy attenuation mode and the meshing sideband broadening characteristics together carry the time-frequency information of local damage in the transmission chain.

[0033] It should be noted that the diagnostic sequence includes both vibration-side feature markers and acoustic emission-side feature markers. The former corresponds to the fault cycle time in the bearing raceway envelope that exceeds the amplitude threshold, while the latter corresponds to the timestamp in the gear meshing pulse segment that is determined to be an abnormal transient impact response. The two types of markers are arranged in ascending order and crosswise on the time axis of the same sampling clock reference.

[0034] Specifically, a time window is taken from the diagnostic sequence before and after the timestamp marked by the vibration side characteristics, and the corresponding impact time-domain waveform segment in the raceway purification envelope is extracted; the amplitude decay trajectory after the impact peak point is collected point by point along the time axis, and the amplitude decay trajectory is fed into the least squares fitting process, with the fitting model adopting a one-sided exponential decay form. Where A(t) is the instantaneous amplitude of the decay trajectory at time t, A0 is the initial impact amplitude at the moment the impact peak occurs, τ is the decay time constant, and t is timed from zero at the moment of the impact peak. By traversing the candidate values ​​of τ and A0, the sum of squared residuals of the fitting is minimized, and the τ and A0 that minimize the sum of squared residuals are read as the two components of the impact energy decay morphology.

[0035] In one possible implementation, the average decay time constant collected during the equipment break-in period is taken as the baseline τ0, and the ratio of τ to τ0 under the current operating conditions is denoted as... The following intervals can be used to classify the impact energy absorption characteristics of the bearing structure: When k is between 0.8 and 1.2, it is determined that the absorption characteristics of the bearing structure have not changed significantly, the raceway is in normal operation, and routine periodic inspections are sufficient; When k is between 1.2 and 1.35, the dissipation rate of impact energy has slightly decreased, indicating that initial micro-pitting corrosion has appeared on the raceway surface. It is recommended to shorten the vibration and temperature sampling interval from 24 hours to 8 hours and observe for 3 consecutive sampling cycles to confirm the trend; When k is between 1.35 and 1.5, it is determined that the pitting corrosion has entered the expansion stage but has not yet penetrated to form a macroscopic peeling layer. Lubricating oil ferrography analysis and endoscopic verification should be arranged during the next shutdown window; When k exceeds 1.5, it is determined that the residence time of impact energy in the bearing structure has significantly increased, the pitting corrosion of the raceway has expanded to the peeling stage, and a shutdown maintenance procedure should be triggered. To avoid repeated jumps near the critical value, a hysteresis band of 0.02 is introduced at the boundary of each interval. That is, when moving from a low level to a high level, the boundary value is increased by 0.02, and when moving from a high level back to a low level, the boundary value is decreased by 0.02. Furthermore, the condition must be met for two consecutive samples before the stage can be switched.

[0036] Understandably, after localized tooth surface spalling occurs during gear meshing, the meshing stiffness experiences a periodic decrease at the moment the spalled tooth pair engages in meshing. This periodic decrease modulates the meshing vibration at a low frequency, with the modulation frequency equal to the rotational frequency of the faulty gear. Further, the corresponding segments of the gear meshing acoustic emission pulses are extracted from the first diagnostic sequence according to the timestamps marked by the acoustic emission side characteristics; the extracted meshing pulse segments are divided into several overlapping short time windows along the time axis, and a short-time Fourier transform is performed on each short time window after applying a Hanning window to obtain a time-frequency spectrum matrix; the theoretical fundamental frequency fm0 of gear meshing is calculated from the product of gear speed and number of teeth, and the spectral peak with the largest amplitude and frequency deviation not exceeding two percent of fm0 in the neighborhood of fm0 in the time-frequency spectrum matrix is ​​located as the fundamental frequency fm of gear meshing; spectral lines with peak amplitudes exceeding three times the average of background spectral lines are symmetrically searched on both sides of fm and identified as sideband spectral peaks; the frequency offset Δf of each sideband spectral peak relative to fm and the ratio R of the sideband spectral peak amplitude to the spectral peak amplitude at fm are measured; the average of Δf corresponding to the sideband spectral peak with the largest frequency offset and R of all sideband spectral peaks together constitute two components of the meshing sideband broadening feature.

[0037] Preferably, the symmetry of the sideband spectral peaks is checked simultaneously. If sideband spectral peaks with similar amplitudes appear at symmetrical positions on both the left and right sides of fm, it is determined that the measured sideband spectral peaks are caused by tooth surface peeling modulation; if only one side has obvious sideband spectral peaks, it is determined to be background noise interference.

[0038] In one embodiment, the two components of the impact energy attenuation morphology are assigned to the bearing channel, and the two components of the meshing sideband widening characteristic are assigned to the gear channel. During a continuous 72-hour steady-state operation phase after equipment break-in, each component is sampled at 10-minute intervals, and the minimum and maximum values ​​are retained to form historical steady-state baseline data. Range normalization is performed on the current components in both channels, i.e., the current component is subtracted from the minimum value in the historical steady-state baseline data and then divided by the difference between the maximum and minimum values, mapping it to a numerical range between 0 and 1. The normalized bearing channel components are then concatenated in a fixed order, followed by the gear channel components, to form a four-dimensional column vector, thus obtaining the damage symptom vector.

[0039] It is understandable that the damage symptom vector simultaneously carries information on the changes in the impact absorption characteristics of the bearing bearing structure and the changes in the gear meshing stiffness modulation characteristics. These two types of information coexist in the same vector according to channel partitions, providing a unified input basis for subsequent identification of the wear stages of each stage of the transmission chain.

[0040] S103. Based on the damage symptom vector, identify each level of the transmission chain components step by step, identify the current stage of wear and degradation, and determine the spatial coordinates of the fault source in the transmission chain.

[0041] Following the mechanical hierarchy of the drivetrain—first-stage reduction gear shaft, second-stage intermediate shaft, and power output shaft—the bearing and gear channel components in the damage symptom vector are filtered using the rotational frequency and meshing frequency windows of each component level. Components falling within the characteristic frequency windows of each level are assigned to their corresponding levels, resulting in symptom response slices for each component level. For each component's symptom response slice, a support vector machine (SVM) classifier, pre-trained with four operating condition samples (normal break-in, early pitting, moderate spalling, and severe failure), is input. The SVM classifier outputs a label for the operating condition category of each component level, determining the current stage of wear degradation in each component of the drivetrain. From the operating condition category labels of each component level, level numbers deviating from the normal break-in category are filtered out. These level numbers are mapped to the corresponding shaft segment numbers and bearing seat numbers on a pre-established drivetrain assembly drawing, determining the spatial coordinates of the fault source in the drivetrain.

[0042] In the transmission system of injection molding equipment, the power is output from the self-driven motor, which drives the first-stage reduction gear shaft through meshing with the first-stage reduction gear, then drives the second-stage intermediate shaft through meshing with the second-stage gear, and finally drives the screw or mold clamping screw through the power output shaft. Each stage of the shaft is supported at both ends by deep groove ball bearings or tapered roller bearings. The number of teeth and module of each stage of the gears, as well as the number of rolling elements and raceway diameter of each stage of the bearings, are all known assembly parameters. The damage symptom vector is a concatenation of bearing channel components and gear channel components, and needs to be restored to the specific component according to the mechanical hierarchy of the transmission chain.

[0043] It should be noted that each component exhibits a different characteristic frequency under steady-state operation. The rotational frequency *fri* of each shaft segment is calculated based on the rated speed of the drive motor and the transmission ratio of each stage, where *i* takes values ​​from 1 to 3, corresponding to the first-stage reduction gear shaft, the second-stage intermediate shaft, and the power output shaft, respectively. The meshing frequency is obtained by combining *fri* with the number of teeth of each gear. Where zi is the number of teeth of the gear in that stage; the bearing outer ring passes through at a frequency of... Where Ni is the number of rolling elements in this class of bearing, di is the diameter of the rolling element, Di is the pitch circle diameter, and αi is the contact angle.

[0044] Specifically, taking the characteristic frequencies fmi and fbi of each level of components as the center, five percent of fri is expanded to both sides as the width of the frequency window, forming a first-level window, a second-level window, and a third-level window. Each component of the damage symptom vector is traced back to its corresponding frequency band according to the source frequency band index recorded during its generation process. The bearing channel component and the gear channel component falling into the first-level window are assigned to the symptom response slice of the first-level component, the component falling into the second-level window is assigned to the symptom response slice of the second-level component, and the component falling into the third-level window is assigned to the symptom response slice of the third-level component. Each level of symptom response slice is separately spliced ​​into a four-dimensional column vector of that level in the order of bearing channel first and gear channel last.

[0045] In one possible implementation, if no valid component is detected within the window of a certain component level, the symptom response segment for that level is filled with the average value of each component collected during the 72-hour steady-state operation phase after the equipment break-in is completed, to prevent subsequent identification processes from failing due to missing items. Furthermore, the training process of the support vector machine classifier is completed offline, with training samples derived from the historical operating records of the same model of injection molding equipment. The symptom response samples collected during the equipment break-in period were classified as normal break-in samples. Samples collected when the vibration kurtosis index increased by 20% to 50% from the break-in baseline value and the equipment was still in service were classified as early pitting samples. Samples collected when flaking or obvious spalling pits were found on the raceway during disassembly and inspection were classified as moderate spalling samples. Samples collected before and after the equipment stopped and alarmed were classified as severe failure samples. After the four types of samples were sorted into roughly balanced proportions, a one-to-many support vector machine classifier was trained using a radial basis function kernel function. The penalty factor was set to 1, and the kernel width parameter was the reciprocal of the mean variance of each dimension of the training samples. After training, the parameters of each support vector and decision hyperplane were fixed.

[0046] In one embodiment, the four-dimensional column vector of each stage of the transmission chain, formed by splicing the symptom responses of each stage, is sequentially fed into the support vector machine classifier. The support vector machine classifier outputs the working condition category label for each stage component, resulting in an identification result sequence containing three-level category labels. Each level category label in the identification result sequence is bound to the stage component, and the current stage of wear and degradation of the damage symptom vector on each stage component of the transmission chain is determined.

[0047] It is understandable that the identification result sequence of the first-stage reduction gear shaft being in the early pitting stage while the second-stage intermediate shaft and power output shaft are in the normal running-in stage indicates that the damage only originated at the front end of the transmission chain; if both the first-stage reduction gear shaft and the second-stage intermediate shaft are in the moderate spalling stage, it indicates that the damage has extended downstream along the transmission chain.

[0048] Preferably, the level number of the working condition category label deviating from the normal break-in category is selected from the identification result sequence, and the level number is sent to a pre-established transmission chain assembly drawing lookup table. The transmission chain assembly drawing records the sequence number of each shaft segment and the bearing seat numbers at both ends. The shaft segment sequence number and bearing seat number corresponding to the deviation level are read from the lookup table result, and the two are combined as the spatial coordinates of the fault source in the transmission chain, thereby locating the wear degradation to the specific installation position.

[0049] S104. Locate the installation position of the faulty component in the injection molding equipment according to the spatial coordinates, obtain the thermal infrared temperature field distribution and lubricating oil metal abrasive spectrum of the installation position, and assess whether the fault influence domain has covered the key cross section of the injection molding equipment's load-bearing structure.

[0050] Based on the shaft segment number and bearing seat number in the spatial coordinates, the physical installation position corresponding to the spatial coordinates is retrieved from the pre-established digital assembly map of the injection molding equipment. This physical installation position is then determined as the on-site calibration coordinates of the faulty component. An uncooled thermal infrared imager is deployed within the neighborhood of these on-site calibration coordinates to collect the infrared radiation intensity matrix of the faulty component's outer shell surface. This matrix is ​​then converted into a temperature field matrix through blackbody radiation calibration. Simultaneously, an oil sample is extracted from the sampling valve port of the transmission system's lubricating oil circulation loop. The emission spectral intensity of iron, copper, and chromium in the oil sample is measured using an atomic emission spectrometer to obtain the metallic abrasive spectrum of the lubricating oil. For the temperature field matrix, the Otsu threshold method is used to adaptively determine the segmentation threshold of the abnormal temperature rise region, and the isothermal closed contour of the abnormal temperature rise region is extracted as the temperature field diffusion boundary shape. The temperature field diffusion boundary shape is overlaid on the digital map of the final assembly, and the spatial projection areas of the temperature field diffusion boundary shape and the tie rod root section, the elbow hinge pin section, and the frame load-bearing beam section are compared to see if they geometrically intersect, and it is determined whether the fault influence domain has covered the key sections of the injection molding equipment's load-bearing structure.

[0051] In the overall structure of the injection molding equipment, the frame load-bearing beam, as the main load-bearing component for the mold closing force reaction, runs horizontally between the upper and lower mold plates. Four tie rods pass through the fixed and moving mold plates and are fixed at both ends with nuts. The cross-section at the root of the tie rod is the location of the transition fillet where the tie rod connects to the threaded end plate. This cross-section bears the cyclic tensile load during the mold closing process. The toggle hinge pin is located at the toggle hinge position of the toggle mold closing mechanism, and the pin cross-section bears the shear load at the moment of mold closing. These critical cross-sections are highly sensitive to local temperature rise and abrasive contamination under long-term cyclic loads. Once the fault influence area reaches these critical cross-sections, the injection molding clamping accuracy will deviate.

[0052] It should be noted that the pre-established digital assembly map of the injection molding equipment records the spatial geometric positions of all shaft segments, gearboxes, oil lines, and frame load-bearing components of the entire machine using a 3D solid modeling method. Each entity has a unique number, and each number is correlated with the shaft segment number and bearing seat number on the transmission chain assembly drawing. The map also pre-sets the spatial projection areas of the tie rod root section, the elbow hinge pin section, and the frame load-bearing beam section, and pre-sets the axial heat transfer path line and oil flow path line along the surface of the transmission shaft housing and the wall of the oil line as a comparison benchmark for the overlap of influence domains and the reading of conduction distance.

[0053] Specifically, the shaft segment number and bearing seat number are extracted from the spatial coordinates, and used as a joint index key to enter the assembly digital map for table lookup, obtaining the three-dimensional coordinates of the uniquely matching physical installation position; along the axial heat transfer path line and the oil flow path line preset in the map, the straight-line distances L1, L2, L3 from the physical installation position to the tie rod root section, the elbow hinge pin section, and the frame load-bearing beam section are read, along with the corresponding conduction path lengths S1, S2, S3. These conduction path lengths are converted into path detour coefficients using k=S / L, used to correct the arrival delay of the subsequent temperature rise baseline; simultaneously, using S1, S2, S3 as attenuation variables, according to... A path attenuation factor is generated, where S0 is the characteristic attenuation length of the material in the axis segment in the spectrum, with a value range of 0.8 meters to 1.2 meters, and the value of a falls between 0.35 and 0.85. It is then used as the row index offset in the penetration depth classification table for table lookup, that is, the longer the path, the lower the classification level, and the shorter the path, the higher the classification level. The corrected path attenuation factor, detour coefficient and straight distance of the three sections are written into the field calibration coordinate data package of the faulty component, and output together with the three-dimensional coordinates of the actual installation position to the temperature rise baseline comparison and penetration depth classification stage.

[0054] In one possible implementation, the lens of an uncooled thermal infrared imager is aimed at the neighborhood of the on-site calibration coordinates, with the field of view covering the surface of the faulty component's casing and simultaneously including the surface of the adjacent load-bearing structure. The imager outputs a radiation infrared intensity matrix frame by frame, where each element corresponds to the infrared radiation intensity of a pixel within the 8-14 micrometer wavelength band, measured in watts per square meter per steradian. The radiation infrared intensity matrix is ​​then used to perform temperature inversion according to Planck's blackbody radiation law, denoted as I=ε*σ*f*T. n+I0, where I is the pixel radiance, T is the pixel absolute temperature in Kelvin, ε is the emissivity of the faulty component's outer shell surface (valued at 0.92), σ is the Stefan-Boltzmann constant, f is the optical response factor of the lens and detector, n is the equivalent power obtained by fitting within the set wavelength band (valued at approximately 4.2), and I0 is the environmental background radiation compensation term. ε, f, n, and I0 are calibrated point-by-point in a blackbody furnace within the range of 233 Kelvin to 673 Kelvin with a step size of 10 Kelvin before the thermal imager leaves the factory. The values ​​are obtained by nonlinear least squares fitting and are then fixed in the equipment. The entire temperature measurement range is divided into three sub-ranges: 233 to 313, 313 to 473, and 473 to 673. Within each sub-range, a fast approximation is used: T = a*I + b, where a and b are Kelvin per watt per square meter per steradian and Kelvin, respectively. These are obtained by least-squares fitting of the blackbody radiation intensity corresponding to the temperature at the endpoint of the sub-range. When the inversion temperature approaches ±5 Kelvin from the interval boundary, a linear weighted transition between adjacent results is used to avoid segmental jumps. The inversion process is executed frame by frame. First, the temperature is roughly calculated from the intensity matrix using a nonlinear relationship. Then, the corresponding a and b are used for refinement based on the sub-range. Each element in the output temperature field matrix represents the absolute temperature at the corresponding location on the surface of the faulty component's outer shell. The inversion error does not exceed ±1.5 Kelvin within the calibration range.

[0055] It should be noted that the lubricating oil circulation loop enters the gearbox's friction pairs from the oil pump outlet along the oil pipe, and then flows back to the oil pan via the return oil pipe. The sampling valve port is located on the return oil pipe, and the oil sample carries the metal abrasive particles generated by the various friction pairs. The extracted oil sample is injected into the electrode gap of the atomic emission spectrometer. Under the excitation of a high-voltage electric arc, the oil sample produces characteristic emission lines. The atomic emission spectrometer reads the intensity values ​​of the emission lines corresponding to iron, copper, and chromium according to the preset wavelength channels, obtaining the lubricating oil metal abrasive particle spectrum containing the intensity of the three channel spectral lines of iron, copper, and chromium.

[0056] It is understandable that the intensity of the iron element spectral line corresponds to the amount of abrasive peeling between the bearing raceway and the tooth surface, the intensity of the copper element spectral line corresponds to the amount of abrasive peeling from the bearing bush or cage, and the intensity of the chromium element spectral line corresponds to the degree of damage to the surface chromium-impregnated layer. The combined information of the intensity of the spectral lines of the three elements provides a data basis for subsequent wear mechanism identification.

[0057] Specifically, the Otsu thresholding method is used to automatically determine the segmentation threshold between the temperature rise anomaly region and the normal temperature background for the temperature field matrix: the Otsu thresholding method traverses all possible thresholds in the temperature field matrix, calculates the maximum value of the inter-class variance after each threshold divides the matrix into two classes of temperature pixels, and takes the threshold that maximizes the inter-class variance as the segmentation threshold; pixels with temperature values ​​exceeding the segmentation threshold constitute a temperature rise anomaly pixel set, the temperature rise anomaly pixel set is divided into several connected sub-regions according to the eight-neighbor connectivity rule, and the connected sub-region with the largest number of pixels and an area ratio of more than 5% is taken as the dominant temperature rise region, and isothermal closed contours are formed by connecting pixels along the outer boundary of the dominant temperature rise region, and the isothermal closed contours are taken as the temperature field diffusion boundary shape.

[0058] In one embodiment, during the equipment installation and commissioning phase, the thermal imager and the final assembly digital map are connected by attaching calibration target points with known spatial coordinates to the surface of the shell of the faulty component, simultaneously recording the coordinates of the target points in the pixel coordinate system of the thermal imager and the coordinates in the three-dimensional coordinate system of the map, and obtaining a pre-calibration mapping matrix through least squares solution; the temperature field diffusion boundary morphology is transformed according to the pre-calibration mapping matrix and projected onto the whole machine three-dimensional space of the final assembly digital map; the three-dimensional region surrounded by the projected temperature field diffusion boundary morphology is geometrically intersected one by one with the spatial projection regions of the pre-set tie rod root section, elbow hinge pin section, and frame load-bearing beam section in the final assembly digital map.

[0059] Preferably, the geometric intersection calculation adopts a combination of axis-aligned bounding box coarse screening and polygon intersection fine calculation: first, the intersection is pre-judged using the minimum bounding boxes of both; if the bounding boxes do not intersect, they are directly determined not to intersect; when the bounding boxes intersect, the edge intersection is calculated on the boundary polygons of both; if a non-empty intersection area is obtained, it is determined that the temperature field diffusion boundary shape has geometrically intersected with the key section, and the fault influence domain has covered the key section; if the intersection calculation results for the tie rod root section, the elbow hinge pin section, and the frame load-bearing beam section are all empty, it is determined that the fault influence domain has not yet covered any key section. In another embodiment, several high-contrast calibration target points are pasted at prominent positions on the surface of the faulty component's outer shell. The spatial three-dimensional coordinates (X,Y,Z) of each target point are precisely measured in advance using the global coordinate system of the assembly digital map (usually with the whole machine reference point as the origin). At the same time, thermal images containing these target points are collected using a thermal imager, and the two-dimensional coordinates (u,v) of each target point in the thermal imager pixel coordinate system are extracted manually or automatically. This constructs a set of corresponding point pairs. Let the number of target points be N (N≥4 and non-coplanar to ensure matrix stability). Then, a homogeneous relationship can be established between the pixel coordinates and 3D coordinates of any target point using a 3×4 projection mapping matrix P. Where λ is the depth scaling factor used to balance the scale on both sides of the equation. Substituting all N sets of corresponding points into the equation yields an overdetermined system of equations. The least squares criterion is used to solve for P, i.e., minimizing the reprojection error: Here, `dehom` represents the conversion from homogeneous coordinates to non-homogeneous pixel coordinates. `||·||` represents the Euclidean norm, i.e., the length of the vector. Here, it's used to calculate distances in a two-dimensional plane, specifically by treating the "predicted pixel coordinates" and "actual pixel coordinates" as a two-dimensional vector and calculating its magnitude, i.e., the straight-line distance between the two points (in pixels). This distance value represents the degree of deviation in the reprojection. `i` (subscript) represents the target point index, i = 1, 2, 3, ..., N. i This represents the horizontal pixel index (i.e., column coordinate) of the i-th target point in the thermal image, increasing from left to right. iThis represents the vertical pixel index (i.e., row coordinate) of the i-th target point in the thermal image, increasing from top to bottom. It is used to traverse all target points involved in calibration and distinguish different measurement pairs. This solution can be initialized using the Direct Linear Transform (DLT) algorithm, and then optimized by Levenberg-Marquardt iteration to obtain the pre-calibration mapping matrix P. Through this matrix, any point (X,Y,Z) in three-dimensional space can be mapped to a pixel (u,v) on the two-dimensional thermal image. In this method, this matrix is ​​obtained by least-squares calibration. It integrates the intrinsic parameters, extrinsic parameters, and distortion correction factors of the thermal imager. After calibration, the thermal imager monitors the temperature field changes on the surface of the faulty component in real time, and extracts the temperature field diffusion boundary morphology through temperature threshold segmentation or gradient edge detection. This boundary is usually represented by a closed polygon composed of a set of discrete pixels. Subsequently, the inverse calculation from pixel to 3D space is completed according to a unified path: using the coordinates of N target points in the whole machine coordinate system during the calibration stage as samples, the local spatial plane equation a*X+b*Y+c*Z+d=0 is obtained by least squares plane fitting, where the coefficients a, b, c, and d are directly given by the fitting, and the fitting residual is controlled within 0.5mm. If the residual exceeds the limit, the surface of the component is divided into pieces according to curvature, and a plane is fitted separately for each piece. The boundary of the piece is switched with the curvature radius being less than 50mm. For each pixel point (u,v) on the boundary, the direction vector of the back projection ray passing through the optical center is first solved by the mapping matrix P. Then, the plane equation corresponding to the piece to which the pixel point belongs is solved simultaneously, and the unique intersection point is the projection coordinate of the pixel in the whole machine 3D space. The piece assignment is determined by the nearest neighbor search within the convex hull of the calibration target point. When a faulty component exhibits a significant thickness difference along the ray direction, such as an outer shell thickness greater than 10mm or the presence of a stepped structure, a linear interpolation correction is performed along the ray direction based on the actual wall thickness at that location in the component's digital map. The interpolation weight is set to 0.5 times the wall thickness, ensuring the back-calculated point falls on the inner surface of the shell. This completes the coordinate transformation of the temperature field diffusion boundary from the image plane to the three-dimensional space of the entire assembly digital map, resulting in a closed curved surface or spatial polygonal ring in three-dimensional space. The projected temperature field diffusion boundary is actually a closed three-dimensional region. If the boundary is a planar ring, it encloses a surface region; if it is a curved surface, it needs to be triangulated to form a volume region, representing the spatial range affected by the current fault's thermal influence. To determine whether the fault's influence region endangers critical load-bearing components in the structure, the system needs to perform geometric intersection checks on it one by one with three pre-set key sections in the digital map—the tie rod root section, the elbow hinge pin section, and the frame load-bearing beam section. These key sections are all clearly defined bounded planar polygons in three-dimensional space, and their respective spatial projection regions (i.e., the areas they occupy in space) are known.

[0060] To improve computational efficiency, the intersection operation employs a two-stage strategy combining coarse screening with axis-aligned bounding boxes and fine calculation of polygon intersection. In the first stage, the minimum circumscribed axis-aligned bounding box of the temperature diffusion boundary region and the axis-aligned bounding box of the current key section polygon are calculated separately. The system checks for overlap in the X, Y, and Z dimensions. If any dimension satisfies the condition of no intersection, the two are directly determined not to intersect, and the next stage is unnecessary. Only when all three dimensions have overlapping intervals is the second stage of fine calculation initiated. In the second stage, the temperature diffusion boundary is treated as a spatial triangular mesh surface, and the key section is treated as a planar polygon. The intersection problem is transformed into finding the intersection line between the spatial plane and the triangular mesh, and further checking whether this intersection line falls inside the section polygon. A more commonly used simplified approach is to project the temperature diffusion boundary onto a local two-dimensional plane containing the key section, obtaining a projected polygon. Then, a standard polygon clipping algorithm (such as Sutherland-Hodgman or Weiler-Atherton) is executed within this two-dimensional plane to find the intersection region between the projected polygon and the section polygon. If the area of ​​the intersection region is greater than a set small threshold (considering numerical errors), then the two are determined to have an actual geometric intersection. The mathematical essence of this precise calculation process is to solve for the non-emptiness of the intersection of two sets of points, that is, to determine whether there exists a point that simultaneously satisfies the conditions of belonging to the closed region defined by the temperature field diffusion boundary and belonging to the key section polygon region. In specific calculations, this can be accomplished by finding the intersection of edges: sequentially checking whether each edge of the temperature field boundary polygon intersects with each edge of the key section polygon in three-dimensional space. If there exists any pair of edge intersection points whose distance to the respective planes of the two polygons is less than the tolerance and the intersection points are simultaneously located within the parameter range of the two line segments, then the intersection is confirmed to be non-empty; if no edge intersects, it is also necessary to check whether one polygon completely contains the other polygon (the inclusion relationship is also considered an intersection cover), which is usually verified by the positional relationship between the point and the polygon (such as the ray method). Only when the intersection detection of each of the three key sections with the temperature field diffusion boundary shape returns an empty intersection (i.e., no edge intersection and no inclusion relationship), does the system determine that the current fault temperature field diffusion influence domain has not yet covered any key load-bearing section.

[0061] One method for determining whether the current wear degradation belongs to the normal break-in stage or the abnormal damage stage is as follows: obtain the baseline amplitude of the temperature rise when the frictional heat is conducted along the drive chain axially to each key section under normal break-in conditions, extract the offset of the measured temperature rise amplitude of each key section under the current fault condition that exceeds the baseline amplitude, identify the penetration depth of the offset, and determine whether the current wear degradation belongs to the normal break-in stage or the abnormal damage stage.

[0062] Specifically, from the pre-established normal break-in condition file, the steady-state temperature records of the tie rod root section, the elbow hinge pin section, and the frame load-bearing beam section are read when the frictional heat of the transmission chain reaches thermal equilibrium along the axial direction. The steady-state temperature rise value of each section is obtained by subtracting the ambient temperature during the same period from the steady-state temperature of each section. The average steady-state temperature rise value during the continuous operation phase of the break-in period is taken as the baseline amplitude of the temperature rise of that section. From the temperature field matrix under the current fault condition, the arithmetic mean of the pixel temperature in the spatial projection area of ​​each of the tie rod root section, the elbow hinge pin section, and the frame load-bearing beam section is read, and the ambient temperature during the same period is subtracted to obtain the measured temperature rise amplitude value of each section. The measured temperature rise amplitude value is subtracted one by one from the corresponding temperature rise baseline amplitude value of the section to obtain the temperature rise offset of each key section. Based on the correlation between the thermal penetration depth of the metal load-bearing structure and the cross-sectional temperature rise, the temperature rise offset is pre-divided into three levels: shallow penetration, medium penetration, and deep penetration. The temperature rise offset is then queried to determine the penetration depth of the corresponding cross-section. If the temperature rise offset falls into the shallow penetration level, the current wear and degradation is determined to be in the normal break-in stage. If the temperature rise offset falls into the medium or deep penetration level, the current wear and degradation is determined to be in the abnormal damage stage.

[0063] In the transmission system of injection molding equipment, the frictional heat generated at the friction pair of the faulty component is gradually conducted along the drive shaft housing, gearbox housing, and frame connecting stiffeners to the tie rod root section, the elbow hinge pin section, and the frame load-bearing beam section. When heat is conducted axially in the metal load-bearing structure, its penetration depth depends on the heat source power, the material's thermal diffusivity, and the conduction path length. During the normal break-in phase, the frictional heat power is at a stable low level, and heat only causes a temperature rise in the near-surface area of ​​the load-bearing structure. During the abnormal failure phase, the frictional heat power increases significantly, and heat penetrates into the deeper areas of the load-bearing structure.

[0064] It should be noted that the pre-established normal break-in condition file is created during the 72-hour stable operation phase after the equipment has completed factory commissioning. The file records the surface temperature sequence of the tie rod root section, the elbow hinge pin section, and the frame load-bearing beam section, which are collected once per minute during this phase. At the same time, the ambient temperature sequence of the workshop during this phase is also recorded.

[0065] Specifically, the ambient temperature at the same moment is subtracted from the cross-sectional surface temperature collected every minute in the archive to obtain the steady-state temperature rise value of the cross-section at that moment; the arithmetic mean of the steady-state temperature rise values ​​over 72 hours is taken as the temperature rise baseline amplitude ΔTb of the cross-section; the above statistics are repeated for the tie rod root section, the elbow hinge pin section and the frame load-bearing beam section to obtain the temperature rise baseline amplitude of each of the three key sections.

[0066] In one possible implementation, after the temperature field matrix under the current fault condition undergoes coordinate transformation according to the pre-calibrated mapping matrix, the projection areas of each key section in the three-dimensional space of the map are determined. The temperature values ​​of pixels falling within the projection areas of each section are extracted from the temperature field matrix. The mean μ and standard deviation σ of the extracted pixel temperature values ​​are calculated. Pixel temperature values ​​deviating from μ by more than 3 times σ are identified as outliers and removed. The arithmetic mean of the remaining pixel temperature values ​​is taken to obtain the measured surface temperature of the section. The ambient temperature collected simultaneously on-site is then subtracted to obtain the measured temperature rise amplitude ΔTm of the section. Further, the measured temperature rise amplitude ΔTm is subtracted one by one from the temperature rise baseline amplitude ΔTb of the corresponding section to obtain the temperature rise offset ΔT = ΔTm - ΔTb of the section. This calculation is repeated on the tie rod root section, the elbow hinge pin section, and the frame load-bearing beam section to obtain an offset set containing the temperature rise offsets of the three sections.

[0067] Understandably, in a one-dimensional steady-state heat conduction system within a metal load-bearing structure, the temperature rise in the depth direction decreases exponentially with increasing depth. The temperature rise at depth d corresponds to a monotonically increasing relationship with the surface temperature rise amplitude. The larger the surface temperature rise offset ΔT, the higher the absolute temperature rise corresponding to the same attenuation ratio, and the deeper the heat penetration depth in the load-bearing structure. By discretizing the relationship between ΔT and penetration depth into three levels, the degree of heat intrusion in the load-bearing structure can be directly inferred from the surface temperature rise offset.

[0068] Specifically, the temperature penetration levels are determined based on the thermal permeation characteristics of the load-bearing structural materials. The thermal permeation area from the surface of the cross-section material to 0.2 times the cross-section thickness is defined as the shallow penetration level; the thermal permeation area from 0.2 to 0.5 times the cross-section thickness is defined as the medium penetration level; and the thermal permeation area exceeding 0.5 times the cross-section thickness is defined as the deep penetration level. The inverse calculation of the temperature rise offset from these depth levels is based on the finite element simulation data of thermal conduction provided with the equipment at the factory. The simulation is established under the following conditions: the load-bearing structure uses a steel cross-section with a thermal conductivity of 45 W / (m·K), a specific heat capacity of 480 J / (kg·K), and a density of 7850 kg / m³; the cross-section thickness is taken as the actual value of the equipment, 20 mm; the outer wall is set as a natural convection boundary with a convection heat transfer coefficient of 10 W / (m²·K); the ambient reference temperature is 25 degrees Celsius; the heat source is loaded with steady-state heat flux density under the rated operating conditions of the equipment; and the power values ​​cover three operating points: 0.5 times, 1.0 times, and 1.5 times the rated value. The reverse engineering process is as follows: First, extract the temperature distribution curve along the cross-sectional normal in the simulation model. The position where the temperature rises by 1 degree Celsius compared to the environmental reference value is recorded as the heat penetration front. Then, read the outer wall temperature rise values ​​corresponding to 0.2 times and 0.5 times the thickness of this front, and take a weighted average for the three power conditions to cover operational fluctuations. This yields an outer wall temperature rise offset ΔT of less than 5 degrees Celsius for shallow penetration, between 5 and 15 degrees Celsius for medium penetration, and greater than 15 degrees Celsius for deep penetration. If the actual material or thickness of the equipment differs from the above values, the front position is re-extracted using the same process, and the boundary values ​​of the three intervals are updated.

[0069] For example, a temperature rise offset level classification table is formed by defining ΔT less than 5 degrees Celsius as shallow penetration level, ΔT between 5 and 15 degrees Celsius as medium penetration level, and ΔT greater than 15 degrees Celsius as deep penetration level.

[0070] In one embodiment, the temperature rise offset ΔT of each section in the offset set is sent to the temperature rise offset level table for lookup to obtain the penetration depth level of the section; if the penetration depth levels of the three key sections all fall into the shallow penetration level, it is determined that the current wear and degradation belongs to the normal break-in stage and the frictional heat is within the design allowable range.

[0071] Preferably, if the penetration depth of any one of the three key sections falls into the medium penetration level or the deep penetration level, it is determined that the current wear and degradation is in an abnormal damage stage, and the frictional heat of the faulty component has penetrated into the deep area of ​​the load-bearing structure.

[0072] S105. When the assessment shows that the fault impact area has covered the critical section, identify the wear mechanism type of the faulty component and determine the remaining service life level of the faulty component.

[0073] Under the premise that the fault influence domain has covered the critical cross section, the extracted oil sample is sent to a ferrometer for magnetic field deposition. The ferromagnetic abrasive particles in the oil sample are deposited on the surface of the ferrogram slide in a stepwise manner according to size and magnetic susceptibility. The ferrogram slide is scanned and imaged using an optical microscope in a dual-light source mode of transmission and reflection, resulting in a set of ferrogram particle microscopic images that carry the contours and internal textures of each ferrogram particle. For each particle in the ferrogram particle microscopic image set, four morphological descriptive quantities are extracted: aspect ratio, edge sharpness, surface texture roughness, and thickness shadow distribution. These four morphological descriptive quantities are input into a convolutional neural network classifier that has been pre-trained with three typical ferrogram particle samples of cutting wear, fatigue spalling, and adhesive wear. The class probability vector output by the convolutional neural network classifier for each particle is read, and the class with the highest proportion in the ferrogram particle microscopic image set is statistically analyzed according to the proportion of particles, which is taken as the wear mechanism type of the faulty component. Based on the wear mechanism type and the penetration depth level, a pre-established dual-entry life level mapping table of wear mechanism and penetration depth is queried: the life level mapping table uses cutting wear, fatigue spalling, and adhesive wear as row indexes, and shallow penetration level, medium penetration level, and deep penetration level as column indexes, and the table is filled with three levels of remaining operating life level: sufficient margin, tight margin, and near failure; the level corresponding to the intersection of the wear mechanism type and the penetration depth level in the life level mapping table is read to determine the remaining operating life level of the faulty component.

[0074] In the lubricating oil circulation loop of the injection molding equipment transmission system, metal abrasive particles generated by faulty components during abnormal wear enter the oil sampling valve port with the lubricating oil. The morphology of the abrasive particles is closely related to their formation mechanism. Abrasive particles generated on the raceway or tooth surface under the cutting action of hard micro-protrusions are slender strips; abrasive particles formed by fatigue cracks initiating from the subsurface and spalling under cyclic contact stress are plate-like blocks; abrasive particles generated by local melting and tearing of the metal surface under instantaneous high temperature and high pressure are surface-torn adhesive clumps. The wear mechanisms corresponding to these three types of abrasive particles are cutting wear, fatigue spalling, and adhesive wear, respectively.

[0075] It should be noted that the ferrography instrument uses a high gradient magnetic field to deposit ferromagnetic abrasive particles in the lubricating oil in a stepwise manner from large to small according to their size and magnetic susceptibility along the length of the ferrography plate. Larger abrasive particles are deposited at the entrance end of the ferrography plate, while smaller abrasive particles are deposited at the far end, forming an orderly arranged abrasive particle deposition band.

[0076] Specifically, an optical microscope uses a transmitted light source to penetrate the abrasive grains to observe their internal texture and thickness shadows, while simultaneously using a reflected light source to illuminate the surface of the abrasive grains to obtain surface details; after simultaneous exposure of the two light sources, they are combined into a single microscopic image with the same field of view; the ferrographic slides are scanned segment by segment along the deposition zone and stitched together to obtain a set of ferrographic grain microscopic images that carry the outline and internal texture of each ferrographic grain.

[0077] In one possible implementation, morphological descriptive parameters are extracted for each particle in the set of microscopic images of ferrographic particles. The aspect ratio is the ratio of the long side to the short side of the particle's smallest bounding rectangle, reflecting the particle's elongation; edge sharpness is the mean absolute value of the curvature of each pixel along the particle's contour, reflecting the sharpness of the contour's angles; surface texture roughness is the standard deviation of the pixel grayscale values ​​within the particle in the reflected light source image, reflecting the roughness of the surface undulations; the thickness shadow distribution is calculated by converting the particle pixel grayscale values ​​in the transmitted light source image into an equivalent thickness value according to the Lambert-Beer law, and the histogram distribution entropy of this equivalent thickness value is used as the descriptive parameter, reflecting the uniformity of the particle thickness. The conversion process involves taking the average grayscale value of the particle-free blank ferrographic plate region as the incident light intensity reference I0 before each sampling, with a typical value between 220 and 245. Then, standard iron-based films with known thicknesses of 5 μm, 10 μm, and 20 μm are placed in the same optical path, and their transmitted grayscale I is collected and solved by inversely solving the equation t = -ln(I / I0) / k. The absorption coefficient k is fitted using the least squares method, with the value of k for iron-based abrasive particles ranging from approximately 0.06 to 0.09 per μm. Other materials are calibrated separately. The calibration curve is refreshed every 200 samplings as the light source ages. The particle pixel grayscale is converted to the equivalent thickness t using the same formula, and then the histogram entropy is calculated. Furthermore, the strip-shaped abrasive grains produced by cutting wear have a large aspect ratio, low edge sharpness, and smooth surface texture; the flaky abrasive grains produced by fatigue spalling have an aspect ratio close to 1, high edge sharpness, and uniform thickness shadow distribution; the agglomerated abrasive grains produced by adhesive wear have an aspect ratio close to 1 but high surface roughness and large thickness shadow distribution entropy. Although the latter two types of abrasive grains have overlapping ranges in the aspect ratio dimension, combined with edge sharpness, surface roughness, and thickness shadow distribution entropy, the three types of abrasive grains exhibit a separable joint distribution region in the value space formed by the four morphological descriptive quantities.

[0078] Specifically, the training samples for the convolutional neural network classifier were obtained from a library of historical oil sample ferrographic images from the same type of injection molding equipment. Microscopic images of three typical particle types—cutting wear, fatigue spalling, and adhesive wear—along with their four morphological descriptors and manually labeled categories, were used to form the training set. The convolutional neural network employed a three-layer convolutional structure with two fully connected layers. The ends of the convolutional layers were global average pooled to output a one-dimensional convolutional feature vector. The four-dimensional vector composed of the four morphological descriptors was then concatenated with the one-dimensional convolutional feature vector along the channel direction to complete feature concatenation. The concatenation result was fed into the fully connected layer, which was then connected to a flexible maximum output layer with three nodes, outputting the category probability vectors corresponding to the three wear mechanisms. The training process employed iterative stochastic gradient descent with dynamic momentum, and the loss function was multi-class cross-entropy. After training, the parameters of each convolutional kernel and the weights of the fully connected layers were fixed.

[0079] In one embodiment, the image of each particle in the ferrographic particle microscopic image set and the four morphological description quantities are sequentially fed into a trained convolutional neural network classifier, the corresponding category probability vector is read, and the dimension with the maximum value in the probability vector is taken as the category to which the particle belongs; the proportion of the three categories in the ferrographic particle microscopic image set is calculated according to the proportion of the number of particles, and the category with the highest proportion is taken as the wear mechanism type of the faulty component.

[0080] Understandably, the wear mechanism reflects the physical causes of surface damage to the faulty component, while the penetration depth reflects the degree of frictional heat intrusion into the load-bearing structure; both together determine the number of remaining cyclic loads that the faulty component can withstand in its current state. Cutting wear is still in the stage of surface material removal, and the remaining load-bearing capacity is only slightly impaired; fatigue spalling indicates that cumulative damage has occurred on the subsurface, and the load-bearing capacity declines rapidly; adhesive wear, accompanied by the instantaneous high temperature of localized welding, has altered the microstructure of the load-bearing structure material, and the decline in remaining load-bearing capacity is the most severe.

[0081] Specifically, a pre-established dual-entry life level mapping table of wear mechanism and penetration depth is compiled using row and column indexes. The life level mapping table uses cutting wear, fatigue spalling, and adhesive wear as row indexes, and shallow penetration level, medium penetration level, and deep penetration level as column indexes. Each cell in the table is filled with the corresponding combination of remaining operating life level, with three levels in total: sufficient margin, tight margin, and near failure.

[0082] For example, the cross-filling allowance between cutting wear and shallow penetration is sufficient; the cross-filling allowance between cutting wear and medium penetration is tight; the cross-filling allowance between cutting wear and deep penetration is tight; the cross-filling allowance between fatigue spalling and shallow penetration is tight; the cross-filling allowance between fatigue spalling and medium penetration is tight; the cross-filling allowance between fatigue spalling and deep penetration is near failure; the cross-filling allowance between adhesive wear and shallow penetration is tight; the cross-filling allowance between adhesive wear and medium penetration is near failure; and the cross-filling allowance between adhesive wear and deep penetration is near failure.

[0083] In one embodiment, the wear mechanism type obtained above is used as a row index and the penetration depth level is used as a column index to enter the life level mapping table for lookup. The level corresponding to the row and column intersection position is read to determine the remaining operating life level of the faulty component.

[0084] Preferably, the lifespan rating mapping table is statistically initialized at the time of equipment model shipment using the mechanism type of historical disassembly and inspection files and the measured penetration depth level during disassembly and inspection as coordinate pairs. After the rating is filled in, it is fixed as the reference for table lookup.

[0085] S106. After analysis, the remaining operating life level has dropped below the safety threshold. The optimal shutdown window under the current injection molding cycle time is determined, and a replacement instruction with maintenance timing constraints is generated.

[0086] The remaining operating life level is compared with a pre-set safety threshold, which is set as a tight margin level. If the remaining operating life level has reached a tight margin or near-failure level, it is determined that the remaining life has fallen below the safety threshold, triggering the shutdown window selection process. The currently executing injection molding process cycle sequence is collected from the programmable logic controller of the injection molding equipment. The process cycle sequence includes the start and end timestamps of each stage: mold closing, injection holding, cooling, mold opening, ejection, and part removal. The idle time segments between the end of cooling and the next mold closing start in each injection cycle of the process cycle sequence are scanned. The shortest time required to replace the faulty part is obtained by reading the pre-established spare parts replacement time table. The idle time segment with the longest duration that meets the shortest time requirement is selected as the optimal shutdown window. Based on the start and end times of the optimal shutdown window and the spatial coordinates of the faulty component, a replacement instruction message containing a shutdown time field, a component location field, and a replacement procedure sequence field is compiled. The replacement instruction message is then sent to the equipment controller and spare parts scheduling terminal via the fieldbus to obtain a replacement instruction with maintenance sequence constraints.

[0087] In the continuous production process of injection molding equipment, each injection cycle consists of six stages connected in series: mold closing, injection and holding pressure, cooling, mold opening, ejection, and part removal. Each stage is driven by a programmable logic controller (PLC) triggered by timing signals. There is a brief idle period between the completion of the part removal action and the start of the next mold closing cycle, allowing the robot arm to reset and the mold cavity to be cleaned. This idle period is the insertable stop segment between cycles. If the machine is forcibly stopped during the mold closing, injection and holding pressure, or cooling stages, the melt in the mold cavity will solidify, rendering the work-in-process unusable and potentially damaging the mold cavity. Therefore, part removal actions should be strictly locked within the idle period between cycles.

[0088] It should be noted that the remaining service life level is derived from the lookup result of the aforementioned dual-entry service life level mapping table of wear mechanism and penetration depth. There are three levels in total, including sufficient margin, tight margin, and near failure. The preset safety threshold is set to the tight margin level. When the level reaches or falls below this safety threshold, the shutdown window selection process is initiated.

[0089] Specifically, the remaining operating life level is sent to the threshold comparison logic. If the remaining operating life level is still in the sufficient margin level, it is determined that the remaining life is higher than the safety threshold. In this case, the shutdown window selection process is not triggered, and the current status is only written to the operation log. If the remaining operating life level has reached the tight margin level or the near failure level, it is determined that the remaining life has fallen below the safety threshold, and the shutdown window selection process is triggered.

[0090] Understandably, the programmable logic controller (PLC) continuously records the start and end timestamps of each stage of the injection molding cycle. Arranging the timestamps of several consecutive cycles in chronological order yields the process cycle sequence. Further, the PLC reads the currently executing injection molding process cycle sequence. This process cycle sequence uses the cycle number as the primary key, and each cycle includes six sets of timestamps: the start and end timestamps of the mold closing stage, the injection and holding stage, the cooling stage, the mold opening stage, the ejection stage, and the part removal stage. The time difference between the end timestamp of the part removal stage of each adjacent cycle and the start timestamp of the mold closing stage of the next cycle is scanned, and this time difference is recorded as the duration of the idle time segment between adjacent cycles. All idle time segments are arranged from longest to shortest to obtain a candidate stop window sequence.

[0091] In one possible implementation, the shortest replacement time Tm required for the faulty component is retrieved from a pre-established spare parts replacement time table. The time table is indexed by the shaft segment number and bearing seat number of the transmission chain assembly drawing. Each record contains the total time required for the disassembly, assembly, alignment, and trial operation of the corresponding component. Idle time segments with a duration less than Tm in the candidate shutdown window sequence are removed, and the longest idle time segment is selected from the remaining idle time segments. Its start and end timestamps are used as the start and end times of the optimal shutdown window.

[0092] Understandably, the optimal stop window falls within the gap between adjacent cycles, avoiding the load-bearing states such as injection pressure holding and cooling, thus preventing the scrapping of in-process products and damage to the mold cavity.

[0093] Specifically, when compiling a replacement instruction message, three core fields are carried in a fixed format: the shutdown time field is filled with the start and end timestamps of the optimal shutdown window; the component location field is filled with the shaft segment number and bearing seat number in the spatial coordinates of the faulty component; the replacement process sequence field is expanded according to the process items corresponding to the spare parts replacement time table, and the duration values ​​of each process are filled in sequentially: shutdown preparation, bolt removal, component hoisting, new component hoisting, alignment and calibration, and trial operation start-up. The cumulative sum of the duration values ​​of each process does not exceed Tm.

[0094] In one embodiment, the compiled replacement instruction message is packaged in the industrial Ethernet protocol frame format and broadcast to the equipment controller and spare parts dispatch terminal via fieldbus. After receiving the replacement instruction message, the equipment controller writes the start timestamp in the shutdown time field into the shutdown plan register and waits for the clock to advance to the start timestamp to automatically execute the shutdown action. After receiving the replacement instruction message, the spare parts dispatch terminal pre-distributes spare parts and tooling to the vicinity of the field calibration coordinates according to the component location field and the replacement process sequence field.

[0095] Preferably, a verification field is added to the replacement instruction message, and the receiving device controller and spare parts scheduling terminal perform cyclic redundancy verification on the message; after the verification is passed, an acknowledgment frame is sent to the diagnostic master control terminal that issued the replacement instruction message, forming a replacement instruction with maintenance timing constraints.

[0096] S107. Execute the replacement operation of the faulty part of the injection molding equipment according to the replacement instruction, extract the vibration envelope of the bearing raceway after the replacement is completed and after assembly and calibration, and restart to reach the steady-state operation stage. When it is confirmed that the vibration envelope has returned to the preset healthy operation range, restore the injection molding equipment to the normal production rhythm of the intelligent manufacturing production line.

[0097] The injection molding equipment is driven to a stop state according to the stop time field in the part replacement instruction. The on-site assembly robot arm is driven to perform the removal of the faulty part, hoisting of the new part, and centering and calibration operations according to the part position field and the part replacement process sequence field, so as to obtain the transmission system after the assembly and centering calibration is completed. The transmission system is restarted and waited for it to enter the steady-state operation stage. Vibration signals are collected from the bearing housing acceleration sensor in the steady-state operation stage and demodulated by Hilbert transform to extract the vibration envelope shape of the bearing raceway under the operation of the new part. The upper and lower limit envelopes of the amplitude formed by the vibration envelope collected during the equipment break-in stage are retrieved as the preset healthy operation range. The vibration envelope shape of the bearing raceway is compared with the preset healthy operation range point by point. If the vibration envelope shape of the bearing raceway falls within the preset healthy operation range throughout the entire process, it is determined that the fault has been eliminated. A production cycle recovery instruction is issued to the main control terminal of the intelligent manufacturing production line to restore the injection molding equipment to the normal production cycle.

[0098] In one possible implementation, after receiving the part replacement instruction, the diagnostic master control terminal of the injection molding equipment parses the shutdown time field into an absolute clock signal and pushes it to the equipment controller. When the clock advances to the start timestamp of the shutdown time, the equipment controller cuts off the mold closing power and locks the oil pump, and the injection molding equipment enters a shutdown state. The spare parts scheduling terminal pre-sends the new part to the vicinity of the on-site calibration coordinates according to the component position field, and drives the on-site assembly robotic arm to complete the removal of the faulty component, hoisting of the new part, and centering calibration operations according to the duration value of each process in the part replacement process sequence field, thereby obtaining the transmission system after the assembly centering calibration is completed.

[0099] Specifically, after restarting the transmission system, the drive motor speed is allowed to reach the rated speed and be maintained for more than 10 minutes. At the same time, the bearing housing temperature rise rate is less than 0.1 degrees Celsius per minute and the vibration root mean square value fluctuation is less than 5% of its sliding mean. When the above three conditions are met, it is determined to enter the steady-state operation stage. The original vibration signal is collected from the bearing housing acceleration sensor in the steady-state operation stage and demodulated by Hilbert transform to extract the vibration envelope morphology of the bearing raceway under the new component operation.

[0100] In one embodiment, vibration envelope samples collected from the same model of equipment during the break-in period are retrieved. The mean plus three times the standard deviation of each sample envelope at each time point is taken as the upper limit of amplitude, and the larger of the mean minus three times the standard deviation and zero is taken as the lower limit of amplitude, forming a preset healthy operating range. The vibration envelope shape of the bearing raceway is compared point by point with the preset healthy operating range. If the vibration envelope shape of the bearing raceway falls within the preset healthy operating range throughout the entire process, the fault is determined to be eliminated. The diagnostic master control terminal sends a production cycle recovery command to the intelligent manufacturing production line master control terminal. The production line master control terminal re-enters the injection molding equipment into the normal production queue and restores the normal production cycle, thereby completing the online diagnosis of equipment faults.

[0101] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for online fault diagnosis of intelligent manufacturing production line equipment, characterized in that, The method includes: The vibration envelope of the bearing raceway and the acoustic emission pulse of gear meshing of the rotating mechanism are obtained. The speed and power frequency interference data in the vibration envelope of the bearing raceway are stripped away. The fault cycle response embedded in the acoustic emission pulse of gear meshing is identified to obtain a diagnostic sequence. Based on the diagnostic sequence, the impact energy attenuation morphology and meshing sideband widening characteristics are extracted to obtain the damage sign vector; Based on the damage symptom vector, each level of the transmission chain components are identified step by step to determine the current stage of wear and degradation and to determine the spatial coordinates of the fault source in the transmission chain. Based on the spatial coordinates, locate the installation position of the faulty component, obtain the thermal infrared temperature field distribution and lubricating oil metal abrasive spectrum at the installation position, and assess whether the fault influence domain has covered the critical cross section of the injection molding equipment's load-bearing structure. When the assessment shows that the fault impact area has covered the critical section, the wear mechanism type of the faulty component is identified, and the remaining service life level of the faulty component is determined. Analysis shows that the remaining service life level has dropped below the safety threshold. The optimal shutdown window under the current injection molding cycle time is determined, and a part replacement instruction is generated. According to the replacement instruction, the faulty component is replaced. After replacement, assembly, calibration, and restart, the bearing raceway vibration envelope pattern reaches the steady-state operation stage. When it is confirmed that the vibration envelope pattern has returned to the preset healthy operating range, the injection molding equipment is restored to normal production cycle.

2. The method for online fault diagnosis of intelligent manufacturing production line equipment according to claim 1, characterized in that, The process involves acquiring the bearing raceway vibration envelope and gear meshing acoustic emission pulses of the rotating mechanism, stripping the speed and power frequency interference data from the bearing raceway vibration envelope, identifying the fault cycle response embedded in the gear meshing acoustic emission pulses, and obtaining a diagnostic sequence, including: The original vibration signal is demodulated using Hilbert transform to obtain the bearing raceway vibration envelope. The elastic stress wave is bandpass filtered to obtain the gear meshing acoustic emission pulse. The bearing raceway vibration envelope is subjected to adaptive mode decomposition to remove the speed and power frequency interference components, and retain the intrinsic components of the bearing raceway fault characteristic frequencies to obtain the raceway cleanup envelope. The abnormal transient impact response and the fault period of the raceway cleanup envelope are aligned and fused along the same time axis to form a diagnostic sequence.

3. The method for online fault diagnosis of intelligent manufacturing production line equipment according to claim 1, characterized in that, The step of extracting the impact energy attenuation morphology and meshing sideband widening features based on the diagnostic sequence to obtain a damage sign vector includes: The amplitude decay trajectory of the vibration-side impact waveform in the diagnostic sequence is fitted into a one-sided exponential decay curve to obtain the decay time constant and the initial impact amplitude, thus forming the impact energy decay morphology characteristics. The sideband spectral clusters on both sides of the fundamental frequency of the acoustic emission side meshing pulse are located by short-time Fourier transform to obtain the meshing sideband broadening characteristics. The impact energy decay morphology characteristics are assigned to the bearing channel and the meshing sideband broadening characteristics are assigned to the gear channel. After normalization, they are concatenated into a multidimensional column vector to obtain the damage sign vector.

4. The method for online fault diagnosis of intelligent manufacturing production line equipment according to claim 1, characterized in that, The step of identifying each stage of the transmission chain components according to the damage symptom vector, recognizing the current stage of wear and degradation, and determining the spatial coordinates of the fault source in the transmission chain includes: The bearing channel component and gear channel component in the damage symptom vector are filtered by the rotation frequency and meshing frequency window of each level of component to obtain the symptom response slices of each level of component; the symptom response slices are input into a support vector machine classifier trained with four types of working condition samples: normal running-in, early pitting, moderate spalling and severe failure, and the working condition category label of each level of component is output; the level number that deviates from the normal running-in category is mapped to the shaft segment number and bearing seat number on the transmission chain assembly drawing.

5. The method for online fault diagnosis of intelligent manufacturing production line equipment according to claim 4, characterized in that, The process of locating the installation position of the faulty component based on the spatial coordinates, obtaining the thermal infrared temperature field distribution and lubricating oil metal abrasive particle spectrum at the installation position, and assessing whether the fault influence domain has covered the critical cross-section of the injection molding equipment's load-bearing structure includes: Based on the shaft segment number and bearing seat number, the corresponding physical installation position is retrieved from the digital assembly map of the injection molding equipment as the on-site calibration coordinates. Within the neighborhood of the on-site calibration coordinates, a thermal infrared imager is used to collect the radiation infrared intensity matrix, which is then converted into a temperature field matrix through blackbody radiation calibration. Simultaneously, oil samples are extracted from the lubricating oil circulation loop to measure the emission spectral intensity of iron, copper, and chromium elements, obtaining the metal abrasive spectra of the lubricating oil. The temperature field matrix is ​​used to determine the abnormal temperature rise region using the Otsu threshold method, and the isothermal closed contour of the abnormal temperature rise region is extracted as the temperature field diffusion boundary morphology. The temperature field diffusion boundary morphology is geometrically compared with the spatial projection areas of the tie rod root section, the elbow hinge pin section, and the frame load-bearing beam section to determine whether the fault influence domain has covered the critical sections of the injection molding equipment's load-bearing structure.

6. The method for online fault diagnosis of intelligent manufacturing production line equipment according to claim 1, characterized in that, The determination of whether the fault's influence area covers the critical cross-section of the injection molding equipment's load-bearing structure includes: Steady-state temperature rise values ​​at thermal equilibrium are read from the normal break-in condition records for the tie rod root section, the elbow hinge pin section, and the frame load-bearing beam section. The average value during the break-in period is taken as the baseline amplitude of the temperature rise. The measured temperature rise amplitude values ​​of each section are read from the temperature field matrix of the current fault condition. The temperature rise offset is obtained by subtracting the corresponding baseline amplitude value from the measured temperature rise amplitude value of each section. When the temperature rise offset falls into the shallow penetration range, it is determined that the current wear degradation belongs to the normal break-in stage. When it falls into the medium penetration range or the deep penetration range, it is determined that it belongs to the abnormal failure stage.

7. The method for online fault diagnosis of intelligent manufacturing production line equipment according to claim 1, characterized in that, When the assessed fault impact domain covers the critical cross-section, the wear mechanism type of the faulty component is identified, and the remaining service life level of the faulty component is determined, including: The extracted oil sample is deposited under a magnetic field using a ferrometer and then scanned and imaged using a dual-light source optical microscope to obtain a set of ferrographic particle microscopic images. For each particle in the ferrographic particle microscopic image set, four morphological descriptive parameters are extracted: aspect ratio, edge sharpness, surface texture roughness, and thickness shadow distribution. These parameters are then input into a convolutional neural network classifier trained with three types of samples: cutting wear, fatigue spalling, and adhesive wear, to obtain the wear mechanism type. The remaining service life level of the faulty component is determined by querying a life level mapping table based on the wear mechanism type and the penetration depth level.

8. The method for online fault diagnosis of intelligent manufacturing production line equipment according to claim 7, characterized in that, The step of determining the remaining service life level of the faulty component by querying the service life level mapping table based on the wear mechanism type and penetration depth level includes: the service life level mapping table uses cutting wear, fatigue spalling, and adhesive wear as row indexes, and shallow penetration level, medium penetration level, and deep penetration level as column indexes, and the table is filled with three levels of remaining service life level: sufficient margin, tight margin, and near failure; the corresponding remaining service life level is read from the intersection of the wear mechanism type and penetration depth level in the service life level mapping table.

9. The method for online fault diagnosis of intelligent manufacturing production line equipment according to claim 1, characterized in that, The analysis shows that the remaining service life level has dropped below the safety threshold. The optimal shutdown window under the current injection molding cycle time is determined, and a part change instruction is generated, including: The remaining service life level is compared with the safety threshold set as the margin tight level. When the margin tightness level or near failure level is reached, the shutdown window selection process is triggered. The injection molding process cycle sequence containing the start and end timestamps of each stage of mold closing, injection holding, cooling, mold opening, and ejection is collected from the programmable logic controller of the injection molding equipment. The idle time segment from the end of cooling between each injection cycle to the start of the next mold closing is scanned. The longest time segment that satisfies the shortest replacement time for the faulty part is selected as the optimal shutdown window. Based on the start and end times of the optimal shutdown window and the spatial coordinates of the faulty part, a replacement instruction message containing shutdown time field, part location field, and replacement process sequence field is compiled and sent through the fieldbus to obtain the replacement instruction.

10. The method for online fault diagnosis of intelligent manufacturing production line equipment according to claim 1, characterized in that, The process of performing the faulty component replacement operation according to the replacement instruction, extracting the bearing raceway vibration envelope pattern after replacement, assembly, calibration, and restart to reach a steady-state operation stage, and restoring the injection molding equipment to normal production cycle when the vibration envelope pattern is confirmed to have returned to the preset healthy operating range includes: The injection molding equipment is stopped and the assembly robot arm performs the removal of faulty parts, hoisting of new parts, and centering and calibration to obtain a calibrated transmission system. When the transmission system is restarted until it enters a steady-state operation stage, the vibration signal of the bearing housing is collected and demodulated to extract the vibration envelope of the bearing raceway under the operation of the new part. The upper and lower limits of the amplitude envelope during the break-in period are retrieved as a preset healthy operating range. When the vibration envelope of the bearing raceway falls within the preset healthy operating range throughout the process, the fault is determined to be eliminated, and the injection molding equipment is restored to normal production rhythm.

Images