A robot and injection molding machine cooperative failure monitoring method

Abstract

The present application relates to the field of intelligent manufacturing monitoring, especially to a kind of mechanical hand and injection molding machine collaborative fault monitoring method, the method comprises: mechanical hand and injection molding machine collaborative work sensor data acquisition;Based on the dynamic threshold of reconstructive error of auto-encoder and reconstructive error enhancement collaborative fault monitoring;Dynamic threshold is obtained by margin optimization evaluation to the running state of mold and mechanical hand;Reconstructive error optimization factor is obtained by gradual fault analysis and reconstructive error optimization is carried out;Based on the collaborative fault determination of dynamic threshold and optimized reconstructive error;Fault monitoring implementation for mechanical hand and injection molding machine system application, to improve the fault identification precision in the collaborative operation system of mechanical hand and injection molding machine.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent manufacturing monitoring, and in particular to a collaborative fault monitoring method for a robot and an injection molding machine. BACKGROUND

[0002] In modern manufacturing, especially in the production process of precision products such as automobile parts, consumer electronics, medical devices, etc., the close cooperation of robots and injection molding machines is a key link to achieve automated production, improve efficiency and ensure product quality. A typical collaborative process includes: after the injection molding machine completes injection, pressure holding, and cooling, the mold is opened, the robot enters the mold cavity according to the synchronization signal to perform tasks such as picking or placing inserts, and quickly exits the mold safety area after confirming the operation is completed, and then the injection molding machine closes the mold to start the next cycle. This series of actions requires high accuracy in time synchronization, motion trajectory, end effector state (such as gripping force or vacuum level), and the accuracy and reliability of state signal interaction between devices.

[0003] Due to the high speed, precision and complexity of the collaborative process, any small deviation, delay or component failure, such as robot entry / exit timing error, grabbing position offset, grabbing force / vacuum level anomaly, injection molding machine mold opening / closing anomaly or signal synchronization error, etc., can cause serious consequences, including but not limited to product damage, precision mold collision damage, robot body structure damage, unplanned downtime, and even threaten the safety of on-site personnel, causing significant economic losses and safety hazards. Therefore, real-time, accurate and reliable fault monitoring of the collaborative working state of the robot and the injection molding machine is crucial to ensure production safety, stability and efficiency.

[0004] In order to achieve effective collaborative fault monitoring, the prior art has proposed a method of analyzing a large amount of sensor data collected from the robot and the injection molding machine. These data typically include joint angles, speeds, forces / torques, end effector states of the robot, and mold positions, oil pressure / electric drive forces, barrel temperatures, screw positions / speeds, ejector pin states, etc. A common technical means is an anomaly detection algorithm based on machine learning, especially using autoencoders and their variant models. The basic principle of this method is: first, collect a large amount of multi-sensor time series data during normal collaborative operation of the system to train the autoencoder model. Through unsupervised learning, the autoencoder captures the complex nonlinear correlations and temporal dependencies between the sensor data in the normal working mode, and learns a mapping that can reconstruct the normal input data with low error. In the actual monitoring stage, the collaborative working data collected in real time is input into the pre-trained autoencoder, and the difference between the input data and its reconstructed output data, i.e. the reconstruction error (usually using mean square error When the system is operating normally, the input data conforms to the normal pattern learned by the model, and the reconstruction error remains low. However, when a fault or abnormal condition causes the data pattern to deviate from the normal, the autoencoder cannot accurately reconstruct the input, resulting in a significant increase in the reconstruction error. The real-time reconstruction error is compared with a pre-set fixed threshold. When the error exceeds the threshold, the system determines that a coordinated failure has occurred. This method, which relies on the autoencoder reconstruction error and fixed threshold judgment, has found application in industrial process monitoring because it can be trained without faulty samples.

[0005] While the collaborative fault monitoring method based on autoencoder reconstruction error and a fixed threshold provides a viable technical approach, its core reliance on a fixed threshold for fault detection presents significant technical limitations when applied to highly dynamic, precise, and variable collaborative scenarios such as robot and injection molding machines. This makes it difficult to effectively meet the high demands for monitoring accuracy and timeliness in actual production. Specifically, the core issue with this fixed threshold strategy is that its static and one-size-fits-all judgment criteria cannot adaptively address the inherent dynamic characteristics of the collaborative process, the normal fluctuations in operating conditions, and the need to detect early hidden faults. First, a complete robot-injection molding machine collaborative cycle inherently comprises multiple sub-phases with distinct kinematic and dynamic characteristics and potential risks (such as high-speed entry into the mold cavity, precise gripping / placement, high-speed exit, and external waiting). The normal fluctuation range of sensor data and the required sensitivity to abnormal deviations in each phase naturally vary. Therefore, a fixed threshold cannot simultaneously provide optimal sensitivity and specificity across all phases. This can lead to underreporting of minor deviations in high-risk, critical phases and false positives in non-critical phases with larger normal fluctuations. Secondly, the actual production environment is not constant. Factors such as changing molds, adjusting process parameters (such as injection speed, holding pressure, cooling time), using different batches of raw materials, natural wear and aging of equipment, and changes in ambient temperature will cause the "normal" working mode represented by the sensor data to drift slowly or change in steps, thereby causing the reconstructed error baseline level under normal operating conditions to change. Fixed thresholds cannot track this normal operating condition change. When the error baseline rises as a whole, it is easy to cause a large number of false alarms, interfering with normal production. When the baseline drops or a fault occurs within the new operating range, the effectiveness of the detection will be reduced.

[0006] Furthermore, many serious synergistic failures (for example, the accumulation of slight delays in synchronization signals, the slight decline in gripping force caused by the slow leakage of clamp cylinder sealing, the slight drift in positioning accuracy caused by the early wear of joint drive components, etc.) often show a slight and persistent rise in reconstruction error compared to the normal baseline in the early stage. This signal may be below the fixed threshold for a long time and be ignored, allowing the fault to continue to develop, missing the best opportunity for early warning, diagnosis and preventive maintenance, and not being discovered until the fault worsens to cause obvious abnormalities or even shutdown.

[0007] In summary, the existing technology adopts a fixed reconstruction error threshold for judgment, which cannot dynamically adapt to the complexity and variability of synergistic working conditions, and has inherent defects in balancing the false negative rate and false positive rate of fault detection, especially in timely capturing early weak faults, limiting the overall performance and practical value of the monitoring system. SUMMARY

[0008] Therefore, the present application aims to provide a robot and injection molding machine synergistic fault monitoring method to solve the problem that the method of using a fixed reconstruction error threshold for judgment cannot dynamically adapt to the complexity and variability of synergistic working conditions, and has inherent defects in balancing the false negative rate and false positive rate of fault detection, especially in timely capturing early weak faults.

[0009] To achieve the above-mentioned purpose, the technical scheme of the present application is as follows:

[0010] A robot and injection molding machine synergistic fault monitoring method, comprising the following steps:

[0011] Step S1: Collecting synergistic working sensor data between controlled devices;

[0012] Step S2: Obtaining a dynamic threshold by performing margin optimization evaluation on the operating state of the controlled device;

[0013] Step S3: Obtaining a reconstruction error optimization factor by performing gradual fault analysis and performing reconstruction error optimization;

[0014] Step S4: Synergistic fault judgment based on the dynamic threshold and the optimized reconstruction error;

[0015] Step S5: Fault monitoring implementation for robot and injection molding machine system application.

[0016] Further, the step S2 obtains a dynamic threshold by performing margin optimization evaluation on the operating state of the controlled device, specifically comprising:

[0017] Obtaining the position indicators of the mold opening degree of the injection molding machine and the end effector of the robot relative to the dangerous area in the robot and injection molding machine synergistic system;

[0018] Based on the position index, the first risk factor of the space interaction margin of the reaction equipment is obtained in combination with the cooperative relationship between the mold opening degree and the manipulator position.

[0019] The error time series data of the controlled equipment is collected, the short-term error fluctuation and stability are analyzed, and the first stability factor of the reaction operation fluctuation is obtained.

[0020] The first risk factor and the first stability factor are fused and evaluated, the dynamic factor is obtained, and the short-term moving average of the error is weighted and corrected by taking the dynamic factor as the weight, and the dynamic threshold adapting to the running state change of the controlled equipment is obtained.

[0021] Further, based on the position index, the first risk factor of the space interaction margin of the reaction equipment is obtained in combination with the cooperative relationship between the mold opening degree and the manipulator position, specifically including:

[0022] The position index of the mold opening degree of the injection molding machine and the manipulator end relative to the mold interior or dangerous area in the controlled equipment is obtained; the mold opening degree and the normalized manipulator position index are obtained by normalizing the mold opening degree and the position index respectively; the difference between the two is taken as the relative spatial offset, and the first risk factor weight is obtained by nonlinear mapping through the sigmoid function; the first risk factor weight and the normalized mold opening degree are weighted to obtain the first risk factor reflecting the cooperative running space margin of the controlled equipment.

[0023] Further, the error time series data of the controlled equipment is collected, the short-term error fluctuation and stability are analyzed, and the first stability factor of the reaction operation fluctuation is obtained, specifically including:

[0024] The short-term moving standard deviation and the long-term moving standard deviation of the controlled equipment in the set time window are obtained; the short-term moving standard deviation is taken as the numerator, and the sum of the long-term moving standard deviation and a very small positive number is taken as the denominator to construct the error fluctuation ratio; the ratio is input into the hyperbolic tangent function for nonlinear mapping, and the obtained value is taken as the first stability factor reflecting the state fluctuation degree of the controlled equipment.

[0025] Further, the first risk factor and the first stability factor are fused and evaluated, the dynamic factor is obtained, and the short-term moving average of the error is weighted and corrected by taking the dynamic factor as the weight, and the dynamic threshold adapting to the running state change of the controlled equipment is obtained, specifically including:

[0026] Obtaining a first risk factor, a first stability factor, and a set basic standard deviation multiple of the controlled device; adding the first risk factor and the first stability factor as a comprehensive state indicator, and mapping them through a power function with a natural constant as the base to obtain a first dynamic assessment factor; multiplying the first dynamic assessment factor by the basic standard deviation multiple to obtain a dynamic factor;

[0027] The current short-term moving average error and long-term moving standard deviation of the controlled device are obtained, and the long-term moving standard deviation is weighted with the dynamic factor as a weight, and then added to the short-term moving average error to obtain a dynamic threshold reflecting the change in the device operating state.

[0028] Furthermore, step S3 obtains a reconstruction error optimization factor through gradual fault analysis and performs reconstruction error optimization, specifically including:

[0029] Collect error time series data of the controlled device; obtain a first trend factor reflecting the abnormal trend by performing trend change rate analysis on the error time series data;

[0030] Evaluate the cumulative amount of errors in the reconstruction error that exceed the long-term average level to obtain the cumulative excess error; statistically accumulate the error fluctuation amplitude within the set time series window to form the window fluctuation cumulative sum; perform a ratio analysis on the cumulative excess error and the window fluctuation cumulative sum, and obtain the persistent deviation factor through function mapping;

[0031] The first trend factor and the continuous deviation factor are comprehensively evaluated to obtain the reconstruction error optimization factor. The reconstruction error optimization factor is used as the weight to perform weighted correction on the original reconstruction error to obtain the effective reconstruction error that reflects the gradual fault state of the equipment.

[0032] Furthermore, the error time series data of the controlled device is collected; and a first trend factor reflecting an abnormal trend is obtained by performing a trend change rate analysis on the error time series data, specifically including:

[0033] The short-term moving average error change rate of the controlled device is collected, and its corresponding long-term historical moving standard deviation is obtained; a fraction is constructed with the error change rate as the numerator and the sum of the long-term standard deviation and a very small positive number as the denominator, and the fraction is mapped by a power function with a natural constant as the base to obtain a trend assessment factor; the trend assessment factor is added to a constant 1, and then mapped by a natural logarithm function to obtain a first trend factor reflecting the stability of the error change.

[0034] Furthermore, the cumulative amount of errors exceeding the long-term average level in the evaluated reconstruction error is used to obtain the cumulative excess error; within the set time series window, the error fluctuation amplitude is statistically accumulated to form the window fluctuation cumulative sum; the cumulative excess error is analyzed by ratio with the window fluctuation cumulative sum, and the persistent deviation factor is obtained through function mapping, specifically including:

[0035] Based on the set window length, the error time series data of the controlled equipment is divided into multiple time series windows; for the error time series data in each window, the difference between it and the corresponding long-term moving average is calculated and compared with the constant 0, and the larger value is taken as the basic measure of excess error; each basic measure is accumulated to obtain the cumulative excess error of the window; at the same time, the long-term moving standard deviation of all error time series data in the window is obtained and accumulated to form the cumulative sum of window fluctuations; a fraction is constructed with the cumulative excess error as the numerator and the sum of the window length multiplied by a very small positive number and the cumulative sum of fluctuations as the denominator, and mapped through the hyperbolic tangent function to obtain the continuous deviation factor reflecting the degree of abnormal deviation of the operating status.

[0036] Furthermore, the first trend factor and the continuous deviation factor are comprehensively evaluated to obtain a reconstruction error optimization factor, and the reconstruction error optimization factor is used as a weight to perform weighted correction on the original reconstruction error to obtain an effective reconstruction error that reflects the gradual fault state of the equipment, specifically including:

[0037] The first trend factor and the continuous deviation factor are respectively used as input indicators, and the square root of the sum of their squares is calculated to obtain the reconstruction error optimization factor reflecting the comprehensive abnormality degree; the optimization factor is input into a power function with a natural constant as the base to obtain the optimization weight after nonlinear mapping; the original reconstruction error of the controlled equipment is weighted by the optimization weight to obtain the effective reconstruction error reflecting the gradual fault state of the equipment.

[0038] Furthermore, step S4 is based on the collaborative fault determination of the dynamic threshold and the optimized reconstruction error, specifically including:

[0039] Obtain the dynamic threshold at the current moment and the corresponding effective reconstruction error; compare the two. When the effective reconstruction error is greater than the dynamic threshold, it is determined that the controlled device is in a coordinated fault or indicates an abnormal state, and the system generates a fault signal and triggers a preset warning response; when the effective reconstruction error is less than or equal to the dynamic threshold, it is determined that the system operation status is normal, maintains the current control state, and no intervention action needs to be triggered.

[0040] Compared with the prior art, the present invention has the following advantages:

[0041] The proposed method for collaborative fault monitoring of a manipulator and injection molding machine effectively addresses the challenges of real-time monitoring of the collaborative operation of the manipulator and injection molding machine and the difficulty identifying hidden faults in existing automated injection molding production processes. This method integrates sensor data acquisition, autoencoder reconstruction error analysis, dynamic threshold calculation, and a trend fluctuation optimization mechanism to create a highly sensitive and robust fault monitoring framework for industrial sites. By incorporating a margin assessment of the relationship between the injection molding machine mold opening and the manipulator's end position, the risk level during collaborative operation is quantified. Furthermore, through stability analysis, risk and stability factors are extracted from the system's operation, and dynamic factors are generated and used to construct dynamic thresholds. This effectively avoids the problem of static thresholds being unable to adapt to complex operating conditions. Furthermore, the system incorporates a comprehensive analysis of the trend and volatility of the reconstruction error. By evaluating the cumulative excess error and window fluctuation of error time series data, a fusion optimization mechanism for the trend factor and the persistent deviation factor is established. This results in a reconstruction error optimization factor and performs weighted optimization, enhancing the ability to identify gradual, slow-onset, and systemic potential faults. Finally, by jointly judging the optimized reconstruction error and the dynamic threshold, accurate identification and early warning response of the fault status of the collaborative system of the robot and injection molding machine were achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings, which constitute part of the present invention, are provided to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are provided to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0043] Figure 1 This is a flow chart of a method for collaborative fault monitoring of a robot and an injection molding machine according to an embodiment of the present invention. DETAILED DESCRIPTION

[0044] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0045] In the description of the present invention, it should be noted that the terms "upper," "lower," "inner," and "back" and other terms indicating orientations or positional relationships are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate and simplify the description of the present invention. They are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation, and therefore should not be construed as limitations on the present invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0046] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0047] See also Figure 1, is a method flow chart of a method for collaborative fault monitoring of a manipulator and an injection molding machine provided in the first embodiment of the present invention, such as Figure 1 As shown, a collaborative fault monitoring method for a robot and an injection molding machine may include:

[0048] S1, collects collaborative sensor data between controlled devices.

[0049] The method uses a real-time autoencoder reconstruction error sequence as its core input. This reconstruction error is calculated by reconstructing key sensor data related to the robot and injection molding machine (robot position, speed, and end-point state; injection molding machine mold position, pressure, and synchronization signals) using a pre-trained autoencoder model. It quantifies the degree to which the system's state deviates from its normal mode at each time step.

[0050] In order to implement the dynamic threshold adjustment proposed in the present invention, in addition to the core reconstruction error sequence, the original readings of the mold position sensor and the angle readings of each joint of the robot must be synchronously collected.

[0051] The two types of original state signals collected are pre-processed, and the original readings of the mold position sensor are linearly mapped to interval, where 0 represents complete closure, Represents fully open; Based on the real-time collected joint angle data, the three-dimensional coordinates of the manipulator end effector are calculated through the robot forward kinematics model, and then based on the predefined mold geometric area and safety area boundary, the normalized distance or state identifier of the manipulator end effector's current position relative to the dangerous area is calculated (0 represents the deepest / most dangerous place inside the mold, Indicates outside safe area).

[0052] S2, obtains the dynamic threshold by performing margin optimization evaluation on the operating status of the controlled equipment.

[0053] The basis of the present invention is to use an autoencoder model to reconstruct the multi-sensor time series data collected during the collaborative operation of the robot and injection molding machine, and to detect abnormal conditions by analyzing the reconstruction error. The existing technology generally uses a method to continuously compare the real-time reconstruction error with a fixed threshold value that is pre-set offline or based on historical data statistics to determine the fault. This fixed threshold strategy is not suitable for the complex, dynamic and precise application scenario of robot-injection molding machine collaboration, because the collaborative process of the robot and injection molding machine is not a static and uniform process, but a state with significant time-varying characteristics and working condition dependence. A standard collaborative operation cycle encompasses multiple distinct operational phases. For example, during the phase when the robot enters the mold at high speed and with high precision, the system is extremely sensitive to even the slightest trajectory deviation or timing error, posing a high potential collision risk. Therefore, fault detection requires extremely high sensitivity, meaning a relatively low threshold for abnormality determination. Meanwhile, during phases such as when the robot moves within the safe area outside the mold or waits for the injection molding machine to complete its next cycle, the system tolerates a wider range of normal operational deviations, and the immediate risk of minor abnormalities is lower. Using the low threshold currently set for high-score detection in these phases can easily misinterpret normal fluctuations in sensor signals or environmental interference as faults, resulting in false alarms. Conversely, setting a high, fixed global threshold to reduce false alarms inevitably sacrifices detection sensitivity during critical, high-risk collaborative phases, resulting in missed detection of potentially dangerous early deviations and a failure to provide timely warnings. A fixed threshold cannot dynamically adjust its leniency based on the specific collaborative phase, essentially using a static standard to measure a dynamically changing process.

[0054] Secondly, in addition to the dynamic nature of the cycle, the operating conditions and health of the collaborative system itself can also change over time. In actual production, replacing molds with different structures, weights, or precisions, adjusting injection molding process parameters to accommodate different products or materials, batch-to-batch variations in raw material physical properties, normal wear, aging, or performance drift of equipment (such as robot joints and guide rails, and the injection molding machine's hydraulic and servo systems), and even changes in ambient temperature and humidity can all cause normal variations in the distribution characteristics of sensor data. This can lead to slow drift or step changes in the baseline level and fluctuation range of the autoencoder's reconstruction error, even in the absence of faults. A fixed threshold set under initial operating conditions is no longer applicable after these conditions change. If the error baseline rises overall, a large number of normal operations may be misclassified as faults. If the error baseline drops or its fluctuation range changes, the original threshold becomes less sensitive to actual faults under the new operating conditions. This lack of adaptability to changing operating conditions is another major drawback of the fixed threshold approach, seriously affecting the long-term effectiveness of the monitoring system.

[0055] To overcome the aforementioned issues associated with fixed thresholds and achieve more accurate collaborative fault monitoring, this invention proposes a dynamic threshold adjustment mechanism, rather than a static, unchanging threshold. The fault determination threshold should be a dynamic quantity that changes in real time with time and system status. During the dynamic threshold assessment process, the system risk level and operational stability must be evaluated, and the threshold must be calculated and adjusted in real time.

[0056] First, to adapt the threshold to drift in the reconstruction error baseline caused by changing operating conditions, the present invention uses a short-term moving average error as the basis for the dynamic threshold. This short-term moving average error is calculated by averaging the autoencoder reconstruction error within a recent window, providing a real-time reflection of the error center level under current operating conditions. Using this as a benchmark, the dynamic threshold is essentially set as the current error baseline plus a time-varying safety margin.

[0057] Next, the safety margin is evaluated. The purpose of the safety margin is to reflect the acceptable range of normal error fluctuations within the system under current operating conditions. Because the long-term moving standard deviation can represent the statistical fluctuation characteristics of macroscopic operating conditions over a long period of time, the present invention uses the long-term moving standard deviation as a benchmark for measuring this normal fluctuation range. However, using only a fixed multiple of the long-term moving standard deviation as the safety margin reverts to a fixed prediction strategy, making it impossible to adjust sensitivity based on specific circumstances. Therefore, the present invention evaluates the safety margin based on the risk status of the system during operation.

[0058] The system operation risk level of the robot and injection molding machine is quantified by the mold opening of the injection molding machine and the position of the robot end effector relative to the mold interior or dangerous area. The highest risk state occurs when the mold is open and the robot is deep inside the mold. The calculation formula for the first risk factor of the collaborative operation of the robot and injection molding machine is:

[0059]

[0060] in Indicates that the robot and injection molding machine collaborative operating system The first risk factor at the moment; Indicates that the robot and injection molding machine collaborative operating system The mold opening of the injection molding machine at the moment; Indicates that the robot and injection molding machine collaborative operating system The position indicator of the robot end relative to the mold interior or dangerous area at the moment.

[0061] It should be noted that, in order to capture the risk state, first, the difference between the mold opening degree and the position index of the manipulator end relative to the mold interior or dangerous area is calculated, which is a positive number and has a higher value in a high-risk state, and then the function is used to map the value to the interval, obtaining a preliminary risk indication. Considering that there is only a synergistic collision risk when the mold is actually opened, the mapping output is multiplied by the mold opening degree of the injection molding machine for synergistic risk assessment.

[0062] In order to quantify the stability of the system operation, the present application further introduces the stability evaluation of the system. For the first stability factor of the manipulator and the injection molding machine synergistic operation, the calculation formula is:

[0063]

[0064] wherein, represents the first stability factor of the manipulator and the injection molding machine synergistic operation system at time; represents the short-term error fluctuation of the manipulator and the injection molding machine synergistic operation system, i.e. the short-term moving standard deviation; represents the long-term normal fluctuation level of the manipulator and the injection molding machine synergistic operation system, i.e. the long-term moving standard deviation; represents a very small positive number, which is set to in the present embodiment to prevent the denominator from being ; represents the hyperbolic tangent function, which is used to map the ratio to the interval.

[0065] It should be noted that the first stability factor is based on the statistical characteristics of the reconstructed error. When the short-term error fluctuation is significantly greater than the long-term normal level, it indicates that the system is in an unstable state. For the ratio of the short-term error fluctuation to the long-term error fluctuation, the ratio is mapped to the interval by the hyperbolic tangent function, which is used as the first stability factor. When the system is unstable, the first stability factor is close to , and when the system is stable, the first stability factor is close to .

[0066] After obtaining the first risk factor and the first stability factor of the manipulator and the injection molding machine synergistic operation system, the system dynamic factor can be evaluated by the two, and the safety margin can be dynamically evaluated by the system dynamic factor. The calculation formula of the dynamic factor in the manipulator and the injection molding machine synergistic operation system is:

[0067]

[0068] wherein, represents the dynamic factor of the manipulator and injection molding machine collaborative operation system at time ; represents the base standard deviation multiple; represents the first risk factor of the manipulator and injection molding machine collaborative operation system at time ; represents the first stability factor of the manipulator and injection molding machine collaborative operation system at time ; represents the natural constant .

[0069] It should be noted that the dynamic factor of the manipulator and injection molding machine system operation system at time decreases with the increase of the first risk factor and the first stability factor, and a base standard deviation multiple is introduced as the maximum value of the dynamic factor under the ideal state (the ideal state is that the first risk factor and the first stability factor are both ), which represents the maximum relative margin allowed by the system, and the above formula ensures that the value of the dynamic factor is always positive and dynamically changes within the range of , when the risk or instability increases, smoothly decreases, resulting in the safety margin being tightened; when both are low, the dynamic factor approaches , and the margin is relaxed.

[0070] After obtaining the dynamic factor corresponding to each time of the manipulator and injection molding machine system operation system, the dynamic threshold can be evaluated by the dynamic factor. For the dynamic threshold of the system at time , the calculation formula is:

[0071]

[0072] wherein, represents the dynamic threshold of the manipulator and injection molding machine collaborative operation system at time ; represents the short-term moving average error of the manipulator and injection molding machine collaborative operation system at time ; represents the dynamic factor of the manipulator and injection molding machine collaborative operation system at time ; represents the long-term normal fluctuation level of the manipulator and injection molding machine collaborative operation system, i.e. the long-term moving standard deviation.

[0073] It should be noted that the core of the dynamic threshold lies in the dynamic factor's recognition of actual scenarios. When the system is in a high-risk operation stage or exhibits an unstable state, the dynamic factor takes effect, thereby tightening the safety margin and improving detection sensitivity; when the system is in a low-risk stage and operating stably, the dynamic factor increases and relaxes the margin, thereby reducing the misjudgment of normal fluctuations.

[0074] S3, obtain the reconstruction error optimization factor through gradual fault analysis and perform reconstruction error optimization.

[0075] After obtaining dynamic thresholds from the operating systems of the robot and injection molding machine systems, they can be used to monitor real-time faults during the coordinated operation of the robot and injection molding machine. However, these systems often exhibit early, subtle, and persistent or simple anomalies. These include initial wear of the robot's joint bearings, slight leaks in the pneumatic gripper seals that slowly decrease gripping force, or small delays in the synchronization signals between the injection molding machine and robot that accumulate over time. These anomalies often initially cause only a very small but potentially persistent or slowly increasing increase in the reconstruction error relative to the normal baseline. Therefore, simply comparing the instantaneous or short-term average reconstruction error with the dynamic threshold is insufficient to effectively and timely capture these fault modes. Because the slow increase in reconstruction error relative to the normal baseline exhibited by these anomaly modes may remain within the safety margin calculated by the dynamic threshold for a long period of time, these potential fault signals are often ignored as normal fluctuations, leading to underreporting by the monitoring system. This underreporting prevents the fault from being detected in its early stages, and its continued development ultimately leads to a serious degradation of equipment performance.

[0076] Therefore, the sensitivity of early weak faults is enhanced by using the behavioral characteristics of the error signal in the time dimension, thereby solving the problem of insufficient sensitivity of the dynamic threshold for detecting such early persistent faults.

[0077] First, the trend of the reconstruction error is evaluated to obtain the first trend factor. The calculation formula of the first trend factor is:

[0078]

[0079] in, Indicates the collaborative operation system between the robot and the injection molding machine The first trend factor at the moment; Indicates the collaborative operation system between the robot and the injection molding machine The rate of change of the short-term moving average error at time; express The long-term historical moving standard deviation of Represents a very small positive number, used to prevent the denominator from being In the embodiment of the present invention, ; Represents a constant ; Expressed as a natural constant Logarithmic function with base .

[0080] It should be noted that this factor is based on the rate of change (slope) of the short-term moving average error. Perform calculations, It reflects the direction and speed of change of the recent error baseline. To eliminate the dimension and focus on the relative change intensity, the slope needs to be divided by its long-term historical moving standard deviation. Considering that the main focus is on the upward trend (positive slope) and the trend signal is expected to be amplified smoothly, the Function performs numerical mapping, which is approximately But it is differentiable everywhere, and can gently amplify the positive slope signal while suppressing the impact of the negative slope (downward trend). The value is close to when the error has no obvious upward trend , and increases accordingly when the error shows a significant upward trend.

[0081] Secondly, we continue to quantify the degree to which the error continues to deviate from its normal level. We measure the sustained deviation factor of the reconstruction error by measuring the situation in which the reconstruction error continues to exceed its corresponding long-term average level over a period of time. For the sustained deviation factor of the reconstruction error, we first calculate the excess error of each sampling point in the reconstruction error time series window, and accumulate these excess errors in the window to obtain the cumulative excess error. The calculation formula for the cumulative excess error is:

[0082]

[0083] in, Indicates the collaborative operation system between the robot and the injection molding machine The cumulative excess error at the moment; Indicates the collaborative operation system between the robot and the injection molding machine The first time in the reconstruction error window corresponding to the The reconstruction error of sampling points; represents the long-term moving average of the reconstruction error; Represents the maximum value calculation function; The time series window representing the error time series data, in the embodiment of the present invention, the time length covered by the window is set to Second.

[0084] After obtaining the cumulative excess error, the cumulative excess error is normalized and measured using the cumulative sum of the long-term error standard deviations within the window as a benchmark, thereby evaluating the persistent deviation factor. The calculation formula for the cumulative sum of the long-term error standard deviations within the window is:

[0085]

[0086] in, Represents the cumulative sum of the long-term error standard deviations within the window; It represents the long-term normal fluctuation level of the collaborative operating system between the robot and the injection molding machine, that is, the long-term moving standard deviation; A time series window representing the error time series data.

[0087] After obtaining the sum of the cumulative excess error and the long-term error standard deviation within the window, the persistent deviation factor can be evaluated using these two parts. The calculation formula for the persistent deviation factor is:

[0088]

[0089] in, Indicates the collaborative operation system between the robot and the injection molding machine The continuous deviation factor of the moment; Indicates the collaborative operation system between the robot and the injection molding machine The cumulative excess error at the moment; Represents the cumulative sum of the long-term error standard deviations within the window; The number of data points in the time series window representing the error time series data; Represents a very small positive number, used to prevent the denominator from being .

[0090] It should be noted that after obtaining the sum of the cumulative excess error and the standard deviation of the long-term error within the window, the cumulative excess error is normalized using the cumulative sum of the long-term error standard deviation within the window as a benchmark, thereby reflecting the degree of the actually observed cumulative excess error relative to the expected total normal fluctuation within the window.

[0091] After obtaining the first trend factor and the continuous deviation factor, a comprehensive evaluation can be performed through the first trend factor and the continuous deviation factor to obtain the reconstruction error optimization factor. The calculation formula of the reconstruction error optimization factor is:

[0092]

[0093] in, Indicates the collaborative operation system between the robot and the injection molding machine The reconstruction error optimization factor at time t; In the collaborative operating system between the robot and the injection molding machine the first trend factor of the moment; represents the reconstruction error of the moment in the collaborative operation system of the robot and the injection molding machine; the sustained deviation factor of the moment.

[0094] It should be noted that the formula ensures that the value of is non-negative, and when either or increases, the value of will increase, and when both increase, the growth of is more significant.

[0095] After obtaining the reconstruction error optimization factor, the reconstruction error can be weighted and optimized by the reconstruction error optimization factor to obtain the effective reconstruction error, and the calculation formula of the effective reconstruction error is:

[0096]

[0097] wherein, represents the effective reconstruction error of the moment in the collaborative operation system of the robot and the injection molding machine; represents the reconstruction error of the moment in the collaborative operation system of the robot and the injection molding machine; represents the reconstruction error optimization factor of the moment in the collaborative operation system of the robot and the injection molding machine; represents the natural constant .

[0098] It should be noted that by multiplying with an amplification factor controlled by the reconstruction error optimization factor, an “effective reconstruction error” is obtained, which is essentially a weighted or enhanced processing of the original error signal based on its historical behavior characteristics. For those error signals that have a small value but show a sustained deviation or upward trend, since the value of the corresponding will increase, the effective error calculated will be significantly amplified; and for those random and short error spikes (which do not have persistence or obvious trend, the value of will be small, the amplification degree of the effective error is small, thereby enhancing the weak but persistent signal in the early stage of failure to be detected by the dynamic threshold.

[0099] S4, collaborative fault judgment based on dynamic threshold and optimized reconstruction error.

[0100] ​​​​​​​​​​​​​​After the calculation of the dynamic threshold and the effective reconstruction error is completed, the core fault judgment link of the present invention is implemented. This link is the decision output part of the entire monitoring method. It uses the results of the first two steps of optimization to determine whether the current collaborative working state of the robot and the injection molding machine is normal. The basic logic of this judgment is to compare the error signal after sensitivity enhancement processing. Whether it exceeds the tolerance limit that is dynamically adjusted based on the current system risk and stability This comparison replaces the existing practice of simply comparing the original reconstruction error with a fixed threshold, aiming to achieve more accurate and timely fault detection.

[0101] Calculation process: At each discrete time step The system calculates the current dynamic fault threshold according to the process in step S2.1 , and follow the steps Calculate the effective reconstruction error at the current moment . Then, the core comparison judgment is performed: the calculated effective reconstruction error With dynamic threshold Make a comparison.

[0102] Fault judgment output is based on the following rules:

[0103] like , then it is determined that the system is at the current time step A coordinated failure has occurred or indicates an abnormal state. The monitoring system should generate a fault signal and trigger subsequent alarm actions (such as lighting an alarm light, sounding an audible alarm, and displaying fault information on the monitoring interface) or control instructions (such as sending an emergency stop or pause signal to the main control system to prevent further damage). At the same time, the relevant fault time and status information should be recorded for analysis.

[0104] like , then the system is judged to be in normal state. The monitoring system does not generate a fault signal and continues to process the next time step. data.

[0105] This determination process is executed in real time and continuously throughout the entire period when the robot and the injection molding machine work together, thereby achieving continuous monitoring of the system status.

[0106] S5, fault monitoring implementation for robot and injection molding machine system applications.

[0107] This paper proposes a method for collaborative fault monitoring between a robot and an injection molding machine. By introducing dynamic thresholds and optimizing reconstruction errors, it improves upon existing monitoring methods based on autoencoder reconstruction errors. This method can be more effectively applied to real-time monitoring of the collaborative operation of a robot and an injection molding machine in actual production. The method's core advantages lie in its dynamic adaptability and enhanced sensitivity to early, subtle faults.

[0108] In specific applications, this method dynamically adjusts the fault detection threshold based on the real-time risk level of the robot-injection molding machine collaborative task and the system's operational stability. Tightening the threshold in high-risk or unstable conditions improves detection sensitivity for serious faults such as potential collisions and jams; while loosening the threshold appropriately in low-risk, stable conditions reduces false positives for normal fluctuations. Furthermore, by amplifying the original reconstruction error with a reconstruction error optimization factor to obtain an effective error, the method significantly improves detection of early-stage, gradual faults caused by component wear, leakage, and parameter drift.

[0109] Therefore, the present invention is applied to scenarios where robots and injection molding machines collaborate, such as automated part removal and insert placement. Compared with existing fixed threshold methods, it can detect various fault signs such as loss of coordination, positioning deviation, and abnormal gripping force earlier and more accurately, effectively preventing equipment damage, ensuring product quality, reducing unplanned downtime, and improving the safety and reliability of the entire automated production system.

[0110] 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, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for collaborative fault monitoring of a robot and an injection molding machine, characterized in that: The method comprises the following steps: Step S1: Collecting cooperative sensor data between controlled devices; Step S2: Obtaining a dynamic threshold by performing margin optimization evaluation on the operating status of the controlled device; Step S3: Obtaining a reconstruction error optimization factor through gradual fault analysis and performing reconstruction error optimization; Step S4: Collaborative fault determination based on dynamic threshold and optimized reconstruction error; Step S5: Implementing fault monitoring for robot and injection molding machine system applications; The step S2 obtains a dynamic threshold by performing a margin optimization evaluation on the operating state of the controlled device, specifically comprising: obtaining the position index of the injection molding machine mold opening and the manipulator end effector relative to the dangerous area in the collaborative system of the manipulator and the injection molding machine; based on the position index, combined with the collaborative relationship between the mold opening and the manipulator position, obtaining a first risk factor reflecting the spatial interaction margin of the device; collecting error time series data of the controlled device, analyzing short-term error fluctuations and stability, and obtaining a first stability factor reflecting operation fluctuations; fusing and evaluating the first risk factor and the first stability factor to obtain a dynamic factor, and using the dynamic factor as a weight to perform a weighted correction on the short-term moving average of the error, thereby obtaining a dynamic threshold that adapts to changes in the operating state of the controlled device; The step S3 obtains a reconstruction error optimization factor through gradual fault analysis and performs reconstruction error optimization, specifically including: collecting error time series data of the controlled device; obtaining a first trend factor reflecting the abnormal trend by performing trend change rate analysis on the error time series data; evaluating the cumulative amount of errors exceeding the long-term average level in the reconstruction error to obtain a cumulative excess error; within a set time series window, statistically accumulating the error fluctuation amplitude to form a window fluctuation cumulative sum; performing ratio analysis on the cumulative excess error and the window fluctuation cumulative sum, and obtaining a continuous deviation factor through function mapping; comprehensively evaluating the first trend factor and the continuous deviation factor to obtain a reconstruction error optimization factor, and using the reconstruction error optimization factor as a weight to perform weighted correction on the original reconstruction error to obtain an effective reconstruction error reflecting the gradual fault state of the equipment.

2. The method for collaborative fault monitoring of a robot and an injection molding machine according to claim 1, characterized in that: The method of obtaining the first risk factor of the spatial interaction margin of the reaction equipment based on the position index and the synergistic relationship between the mold opening and the position of the robot includes: The mold opening of the injection molding machine and the position index of the end of the robot relative to the inside of the mold or the dangerous area in the controlled equipment are obtained; the mold opening of the injection molding machine and the position index are normalized respectively to obtain the normalized mold opening and the normalized robot position index; the difference between the two is used as the relative spatial offset, and nonlinear mapping is performed through the sigmoid function to obtain a first risk factor weight; the first risk factor weight and the normalized mold opening are weightedly calculated to obtain a first risk factor reflecting the collaborative operation space margin of the controlled equipment.

3. The method for collaborative fault monitoring of a robot and an injection molding machine according to claim 1, characterized in that: The collecting of error time series data of the controlled device, analyzing short-term error fluctuation and stability, and obtaining a first stability factor reflecting operation fluctuation specifically includes: The short-term moving standard deviation and long-term moving standard deviation of the error time series data of the controlled device within a set time window are obtained; the short-term moving standard deviation is used as the numerator, and the result of summing the long-term moving standard deviation and a very small positive number is used as the denominator to construct the error fluctuation ratio; the ratio is input into the hyperbolic tangent function for nonlinear mapping, and the obtained value is used as the first stability factor reflecting the degree of state fluctuation of the controlled device.

4. The method for collaborative fault monitoring of a robot and an injection molding machine according to claim 1, characterized in that: The step of fusing and evaluating the first risk factor and the first stability factor to obtain a dynamic factor and using the dynamic factor as a weight to perform weighted correction on the short-term moving average of the error to obtain a dynamic threshold adapted to changes in the operating state of the controlled device specifically includes: Obtaining a first risk factor, a first stability factor, and a set basic standard deviation multiple of the controlled device; adding the first risk factor and the first stability factor as a comprehensive state indicator, and mapping them through a power function with a natural constant as the base to obtain a first dynamic assessment factor; multiplying the first dynamic assessment factor by the basic standard deviation multiple to obtain a dynamic factor; The current short-term moving average error and long-term moving standard deviation of the controlled device are obtained, and the long-term moving standard deviation is weighted with the dynamic factor as a weight, and then added to the short-term moving average error to obtain a dynamic threshold reflecting the change in the device operating state.

5. The method for collaborative fault monitoring of a robot and an injection molding machine according to claim 1, characterized in that: The step of collecting error time series data of the controlled device and performing trend change rate analysis on the error time series data to obtain a first trend factor reflecting an abnormal trend specifically includes: The short-term moving average error change rate of the controlled device is collected, and its corresponding long-term historical moving standard deviation is obtained; a fraction is constructed with the error change rate as the numerator and the sum of the long-term standard deviation and a very small positive number as the denominator, and the fraction is mapped by a power function with a natural constant as the base to obtain a trend assessment factor; the trend assessment factor is added to a constant 1, and then mapped by a natural logarithm function to obtain a first trend factor reflecting the stability of the error change.

6. The method for collaborative fault monitoring of a robot and an injection molding machine according to claim 1, characterized in that: The cumulative amount of errors exceeding the long-term average level in the evaluated reconstruction error is obtained to obtain the cumulative excess error; within the set time series window, the error fluctuation amplitude is statistically accumulated to form the window fluctuation accumulation sum; Perform ratio analysis on the cumulative excess error and the accumulated sum of window fluctuations, and obtain the continuous deviation factor through function mapping, including: Based on the set window length, the error time series data of the controlled device is divided into multiple time series windows; For the error time series data in each window, calculate its difference with the corresponding long-term moving average, compare it with the constant 0, and take the larger value as the basic measure of excess error; accumulate each basic measure to obtain the cumulative excess error of the window; obtain the long-term moving standard deviation of all error time series data in the window and accumulate them to form the window fluctuation cumulative sum; construct a fraction with the cumulative excess error as the numerator and the sum of the window length multiplied by a very small positive number and the fluctuation cumulative sum as the denominator, and obtain the continuous deviation factor reflecting the degree of abnormal deviation of the operating status through hyperbolic tangent function mapping.

7. The method for collaborative fault monitoring of a robot and an injection molding machine according to claim 1, characterized in that: The first trend factor and the continuous deviation factor are comprehensively evaluated to obtain a reconstruction error optimization factor, and the original reconstruction error is weightedly corrected using the reconstruction error optimization factor as a weight to obtain an effective reconstruction error that reflects the gradual fault state of the equipment. Specifically, the following steps are performed: The first trend factor and the continuous deviation factor are respectively used as input indicators, and the square root of the sum of their squares is calculated to obtain the reconstruction error optimization factor reflecting the comprehensive abnormality degree; the optimization factor is input into a power function with a natural constant as the base to obtain the optimization weight after nonlinear mapping; the original reconstruction error of the controlled equipment is weighted by the optimization weight to obtain the effective reconstruction error reflecting the gradual fault state of the equipment.

8. The method for collaborative fault monitoring of a robot and an injection molding machine according to claim 1, characterized in that: The collaborative fault determination based on the dynamic threshold and the optimized reconstruction error specifically includes: Obtain the dynamic threshold at the current moment and the corresponding effective reconstruction error; compare the two. When the effective reconstruction error is greater than the dynamic threshold, it is determined that the controlled device is in a coordinated fault or indicates an abnormal state, and the system generates a fault signal and triggers a preset warning response; when the effective reconstruction error is less than or equal to the dynamic threshold, it is determined that the system operation status is normal, maintains the current control state, and no intervention action needs to be triggered.

Images