A method and system for monitoring the state of injection molding equipment

Abstract

The present invention discloses a state monitoring method and system for injection molding equipment, relating to the technical field of mold state monitoring. The method involves obtaining operating data of the injection molding equipment and converting the operating data into a multidimensional feature vector. A vibration data set of a check valve during operation is obtained, and each vibration data set is weightedly fused to obtain one-dimensional data. The one-dimensional data is then reconstructed to obtain vibration feature data. The multidimensional feature vector and the vibration feature data are used as inputs to a fault detection model to obtain a fault detection result, and the check valve state is monitored based on the fault detection result. By integrating the operating data of the injection molding machine and the vibration data of the check valve, a comprehensive feature data set is generated for fault detection, enabling real-time and accurate state monitoring of the injection molding equipment, improving the reliability and maintenance efficiency of the equipment operation, and reducing the failure rate and repair costs.

Description

Technical Field

[0001] The present invention belongs to the technical field of mold status monitoring, and in particular relates to a status monitoring method and system for injection molding equipment. Background Art

[0002] As a core piece of equipment in modern industrial manufacturing, injection molding machines have undergone a remarkable evolution since their industrialization in the mid-20th century, transitioning from hydraulic drive to fully electric and intelligent systems. Early injection molding machines primarily relied on mechanical hydraulic systems, relying on manual experience to adjust parameters, limiting production efficiency and precision. With breakthroughs in servo motor technology and closed-loop control systems, fully electric injection molding machines with high response and low energy consumption have gradually become mainstream.

[0003] Patent No. CN119858287A discloses a method and system for monitoring the operating status of injection molding equipment; it collects the multi-dimensional operating parameter time series of the injection molding equipment at different times, groups the parameters of the same dimension into a data set, and constructs an isolation tree based on each data set; after sorting the tree nodes of the isolation tree according to the data average value, assigns a tree node number to each data item; regards the combination of parameters of each dimension at the same time as a data point, and calculates their degree of grouping by comparing the distribution relationship of any two data points on the isolation tree, and classifies data points with a degree of grouping higher than a set threshold into the same group; calculates the degree of abnormality of the data point; if the degree of abnormality exceeds the set threshold, it is determined to be an abnormal data point and an alarm is issued at the same time.

[0004] In the existing technology, injection molding machines suffer from internal wear and tear due to long-term use. Traditional injection molding equipment fault monitoring has incomplete data, untimely monitoring, and lacks monitoring accuracy, resulting in a decrease in product yield. Summary of the Invention

[0005] The purpose of the present invention is to solve the problem that the internal wear of the injection molding machine due to long-term use, the data in the traditional injection molding equipment fault monitoring is incomplete, the monitoring is not timely, and the monitoring accuracy is lacking, which leads to a decrease in the yield rate of the product, and proposes a status monitoring method and system for injection molding equipment.

[0006] In a first aspect of the present invention, a condition monitoring method for injection molding equipment is first proposed, the method comprising:

[0007] Acquiring operating data of the injection molding equipment and converting the operating data into a multi-dimensional feature vector; the operating data includes: pressure data, torque data and displacement data;

[0008] Acquire a vibration data set of the check valve during operation, perform weighted fusion on each vibration data to obtain one-dimensional data, and reconstruct the one-dimensional data to obtain vibration characteristic data; the vibration data set includes vibration data of multiple sensors;

[0009] The multidimensional feature vector and the vibration feature data are used as inputs of a fault detection model to obtain a fault detection result, and the state of the check valve is monitored according to the fault detection result.

[0010] Optionally, converting the operating data into a multi-dimensional frequency domain feature vector includes:

[0011] Performing fast Fourier transform on the pressure data, torque data, and displacement data in the operation data to obtain a frequency domain complex sequence set; the frequency domain complex sequence set includes: a first frequency domain complex sequence, a second frequency domain complex sequence, and a third frequency domain complex sequence;

[0012] Performing frequency domain feature extraction on the frequency domain complex number sequences in the frequency domain complex number sequence set to obtain a first frequency domain feature vector, a second frequency domain feature vector, and a third frequency domain feature vector, and fusing the first frequency domain feature vector, the second frequency domain feature vector, and the third frequency domain feature vector to obtain a multidimensional frequency domain feature vector;

[0013] Time domain features are extracted from the operating data to obtain a multidimensional time domain feature vector, and the multidimensional frequency domain feature vector and the multidimensional time domain feature vector are spliced ​​in a preset order to obtain a multidimensional feature vector.

[0014] Optionally, weighted fusion is performed on each vibration data to obtain one-dimensional data, including:

[0015] Get the vibration data x of each sensor i (T) and normalize the vibration data to obtain the target vibration data X i (T), generates a symmetric matrix S according to each sensor;

[0016] The correlation coefficient r between the target sensor i and each sensor is calculated based on the symmetric matrix S i (T), the average value is calculated based on the vibration data of each sensor and standard deviation σ i (T), according to the mean and standard deviation σ i (T) Calculate the variation factor;

[0017] According to the correlation coefficient r i (T) and the mutation factor are used to calculate the fusion weight ω i (T), normalize the fusion weight to get the target weight ω i'(T), according to the target weight ω i '(T) For each vibration data x i (T) Perform weighted summation to obtain one-dimensional data.

[0018] Optionally, reconstructing the one-dimensional data to obtain vibration characteristic data includes:

[0019] Obtaining Gaussian white noise and adding the Gaussian white noise to the one-dimensional data to obtain target one-dimensional data, performing empirical mode decomposition on the target one-dimensional data to obtain a plurality of intrinsic mode components, and averaging the intrinsic mode components to obtain a first-order intrinsic mode function;

[0020] A first residual signal is calculated based on the preprocessed vibration data and the first-order intrinsic mode function, and an empirical mode decomposition is performed on the first residual signal to obtain a second-order intrinsic mode function;

[0021] The first residual signal is subtracted according to the second-order intrinsic mode function to obtain the second residual signal, the second residual signal is iteratively decomposed to obtain the N-order intrinsic mode function, and the N-1th residual signal is subtracted according to the N-order intrinsic mode function to obtain the Nth residual signal. When the residual is a monotonic trend term, the Nth residual signal is used as the vibration characteristic data.

[0022] Optionally, performing status monitoring on the check valve according to the fault detection result includes:

[0023] If the multi-dimensional feature vector is determined to be abnormal and the vibration feature data is determined to be normal, it is determined to be a non-check valve fault;

[0024] If the multidimensional feature vector is determined to be normal and the vibration feature data is determined to be abnormal, it is determined that the check valve is early worn;

[0025] If the multi-dimensional feature vector is determined to be abnormal and the vibration feature data is determined to be abnormal, it is determined that the check valve is severely worn.

[0026] In a second aspect of the present invention, a condition monitoring system for injection molding equipment is provided, comprising: an operation data acquisition module, a vibration data acquisition module, and a fault monitoring module:

[0027] The operation data acquisition module is used to acquire the operation data of the injection molding equipment and convert the operation data into a multi-dimensional feature vector; the operation data includes: pressure data, torque data and displacement data;

[0028] The vibration data acquisition module is used to obtain a vibration data set of the check valve during operation, perform weighted fusion on each vibration data to obtain one-dimensional data, and reconstruct the one-dimensional data to obtain vibration characteristic data; the vibration data set includes vibration data of multiple sensors;

[0029] The fault monitoring module is configured to use the multidimensional feature vector and the vibration feature data as inputs of a fault detection model to obtain a fault detection result, and perform status monitoring on the check valve according to the fault detection result.

[0030] Optionally, the operation data acquisition module includes: a frequency domain conversion module, a feature fusion module and a feature splicing module:

[0031] The frequency domain conversion module is configured to perform fast Fourier transform on the pressure data, torque data, and displacement data in the operation data to obtain a frequency domain complex sequence set; the frequency domain complex sequence set includes: a first frequency domain complex sequence, a second frequency domain complex sequence, and a third frequency domain complex sequence;

[0032] The feature fusion module is used to perform frequency domain feature extraction on the frequency domain complex number sequences in the frequency domain complex number sequence set to obtain a first frequency domain feature vector, a second frequency domain feature vector and a third frequency domain feature vector, and fuse the first frequency domain feature vector, the second frequency domain feature vector and the third frequency domain feature vector to obtain a multidimensional frequency domain feature vector;

[0033] The feature splicing module is used to extract time domain features from the operating data to obtain a multidimensional time domain feature vector, and to splice the multidimensional frequency domain feature vector and the multidimensional time domain feature vector in a preset order to obtain a multidimensional feature vector.

[0034] Optionally, the vibration data acquisition module includes: a first execution, a second execution, and a third execution:

[0035] The first execution module is used to obtain the vibration data x of each sensor i (T) and normalize the vibration data to obtain the target vibration data X i (T), generates a symmetric matrix S according to each sensor;

[0036] The second execution module is used to calculate the correlation coefficient r between the target sensor i and each sensor according to the symmetric matrix S i (T), the average value is calculated based on the vibration data of each sensor and standard deviation σ i (T), according to the mean and standard deviation σ i (T) Calculate the variation factor;

[0037] The third execution module is used to calculate the correlation coefficient r i (T) and the mutation factor are used to calculate the fusion weight ω i (T), normalize the fusion weight to get the target weight ω i '(T), according to the target weight ω i '(T) For each vibration data x i (T) Perform weighted summation to obtain one-dimensional data.

[0038] Optionally, the vibration data acquisition module further includes: a modal decomposition module, a primary decomposition module and an iterative decomposition module:

[0039] The modal decomposition module is used to obtain Gaussian white noise and add the Gaussian white noise to the one-dimensional data to obtain target one-dimensional data, obtain multiple intrinsic modal components by empirical mode decomposition of the target one-dimensional data, and average the intrinsic modal components to obtain a first-order intrinsic mode function;

[0040] The primary decomposition module is used to calculate a first residual signal based on the preprocessed vibration data and the first-order intrinsic mode function, and perform empirical mode decomposition on the first residual signal to obtain a second-order intrinsic mode function;

[0041] The iterative decomposition module is used to subtract the first residual signal according to the second-order intrinsic mode function to obtain the second residual signal, iteratively decompose the second residual signal to obtain the N-order intrinsic mode function, subtract the N-1th residual signal according to the N-order intrinsic mode function to obtain the Nth residual signal, until the residual is a monotonic trend term, then the Nth residual signal is used as the vibration characteristic data.

[0042] Optionally, the fault monitoring module includes: a first judgment module, a second judgment module and a third judgment module:

[0043] The first judgment module is configured to determine that the fault is not a check valve fault if the multidimensional feature vector is judged to be abnormal and the vibration feature data is judged to be normal;

[0044] The second judgment module is configured to determine that the check valve is in early stage of wear if the multi-dimensional feature vector is judged to be normal and the vibration feature data is judged to be abnormal;

[0045] The third judgment module is configured to determine that the check valve is severely worn if the multi-dimensional feature vector is judged to be abnormal and the vibration feature data is judged to be abnormal.

[0046] Beneficial effects of the present invention:

[0047] This invention proposes a condition monitoring method for injection molding equipment. The method obtains the operating data of the injection molding equipment and converts it into a multidimensional feature vector. A vibration dataset of the check valve during operation is obtained, and each vibration data set is weighted and fused to obtain one-dimensional data. This one-dimensional data is then reconstructed to obtain vibration signature data. The multidimensional feature vector and vibration signature data are used as inputs to a fault detection model to obtain a fault detection result, and the check valve condition is monitored based on the fault detection result. By integrating the operating data of the injection molding machine and the vibration data of the check valve, a comprehensive feature dataset is generated for fault detection, enabling real-time and accurate condition monitoring of the injection molding equipment. This improves the reliability and maintenance efficiency of the equipment, and reduces the incidence of failures and repair costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The present invention will be further described below with reference to the accompanying drawings.

[0049] Figure 1 A flow chart of a method for monitoring the state of injection molding equipment provided by an embodiment of the present invention;

[0050] Figure 2 A framework diagram of another method system for monitoring the state of injection molding equipment provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention, in conjunction with the accompanying drawings. Obviously, the described embodiments represent only a portion of the embodiments of the present invention, not all of them. The term "and / or" herein simply describes an association relationship between associated objects, indicating that three possible relationships exist. For example, "A" and "B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, references to "first," "second," and so on in the present invention are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined as "first" or "second" may explicitly or implicitly include at least one of these features. Furthermore, the technical solutions of the various embodiments may be combined, but only if they are achievable by a person of ordinary skill in the art. If a combination of technical solutions contradicts or is unachievable, such combination shall be deemed non-existent and outside the scope of protection claimed by the present invention.

[0052] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0053] The embodiment of the present invention provides a method for monitoring the state of injection molding equipment. Figure 1, Figure 1 A flow chart of a method for monitoring the state of injection molding equipment provided in an embodiment of the present invention. The method comprises the following steps:

[0054] S101, obtaining operation data of the injection molding equipment and converting the operation data into a multi-dimensional feature vector;

[0055] S102, obtaining a vibration data set of the check valve during operation, performing weighted fusion on each vibration data to obtain one-dimensional data, and reconstructing the one-dimensional data to obtain vibration characteristic data;

[0056] S103 , using the multidimensional feature vector and the vibration feature data as inputs of a fault detection model to obtain a fault detection result, and performing state monitoring on the check valve according to the fault detection result.

[0057] Operational data includes: pressure data, torque data and displacement data; vibration data set includes vibration data from multiple sensors;

[0058] A condition monitoring method for injection molding equipment provided in an embodiment of the present invention generates a comprehensive feature data set for fault detection by integrating the operating data of the injection molding machine and the vibration data of the check valve, thereby realizing real-time and accurate condition monitoring of the injection molding equipment, improving the reliability of equipment operation and maintenance efficiency, and reducing the failure rate and maintenance costs.

[0059] In one implementation, injection molding equipment (injection molding machines) are primarily divided into plunger-type and screw-type machines. Plunger-type machines: 1. Structural Features: The plunger (similar to a piston) directly pushes the molten material in the barrel into the mold, eliminating the screw's rotating plasticizing process. 2. Check Valve Configuration: Typically, a check valve is not included. Because the molten material is injected into the mold unidirectionally when the plunger advances and the hopper replenishes the material when it retreats, there's no need to prevent the molten material from flowing back. Application Scenarios: Suitable for low-precision, small-batch production, such as simple daily necessities (e.g., bottle caps). ; Screw injection molding machine: 1. Structural features: Plasticization (molten plastic) and injection are achieved through the rotation of the screw, and the screw has the functions of stirring, compressing and conveying the molten material; 2. Check valve configuration: A check valve must be equipped; when the screw rotates backward during the plasticization stage, the molten material needs to enter the storage chamber through the check valve; when the screw moves forward during the injection stage, the check valve is closed to prevent the molten material from flowing back to the back of the screw, ensuring the stability of high-pressure injection; Application scenarios: Widely used in high-precision, large-scale production, such as automotive parts and electronic housings.

[0060] In one implementation, since the injection molding machine's check valve is installed between the barrel and the screw, a sensor cannot be directly installed. Therefore, its vibration characteristics are indirectly obtained through an externally installed vibration sensor. Secondly, the check valve is installed on the screw head, so check valve wear (such as increased radial clearance) will directly lead to abnormal screw movement. Therefore, the screw's operating data indirectly reflects the fault status of the check valve.

[0061] In one implementation, during the use of the check valve, filler particles (such as glass fiber, carbon fiber, calcium carbonate) or carbonized slag in the molten plastic will cut or flush the sealing surface, valve seat, valve needle and other parts of the check valve when the screw rotates and the molten material flows, resulting in scratches, pits or grooves on the sealing surface; the fitting clearance between the valve needle and the valve seat increases (the normal clearance is about 0.01~0.03mm, and may exceed 0.1mm after wear); the valve ball (if it is a ball valve structure) has a worn plane or increased roughness on the surface; the wear of the check valve also includes: corrosion wear (when processing heat-sensitive materials, they decompose at high temperatures to produce corrosive gases, which chemically react with the metal surface of the check valve) and fatigue wear (the check valve is frequently opened and closed during the injection-pressure holding-melting process, and the sealing surface is subjected to cyclic stress, which causes material fatigue and microcracks, which gradually expand into macro wear).

[0062] In one implementation, operating data of the injection molding equipment and vibration data of the check valve during operation are acquired within a preset time period. The check valve data is collected periodically, and the acquisition cycle is within the preset time period. The screw pressure, torque, and displacement data are converted into time-domain data. By extracting the signal's time series statistical features (such as mean, standard deviation, peak value, kurtosis, etc.), the original dynamic signals are converted into quantitative indicators that reflect the equipment's operating status. This effectively reveals subtle changes such as abnormal pressure fluctuations, sudden torque changes, and decreased displacement stability caused by check valve wear. These features can be used as inputs to a machine learning model, improving the accuracy of fault diagnosis while reducing data redundancy and model complexity, enabling real-time monitoring and early warning of check valve wear.

[0063] In one implementation, weighted fusion of multi-channel vibration information suppresses environmental noise interference during the operation of the injection molding machine's check valves, enhancing the signal-to-noise ratio of weak impact features. Dynamic weight allocation optimizes signal fusion quality, improving the concentration of impact energy. This method accurately separates key features such as mechanical motion and fluid impact, improving the accuracy of fault feature frequency identification. Combined with envelope demodulation technology, it can detect early-stage faults such as check valve wear or sticking in real time, reducing the risk of unplanned downtime and extending equipment life. Fault detection models include support vector machines (SVMs) and convolutional neural networks (CNNs).

[0064] In one embodiment, the operating data is converted into a multi-dimensional frequency domain feature vector, including:

[0065] Performing fast Fourier transform on the pressure data, torque data, and displacement data in the operating data to obtain a frequency domain complex sequence set; the frequency domain complex sequence set includes: a first frequency domain complex sequence, a second frequency domain complex sequence, and a third frequency domain complex sequence;

[0066] Performing frequency domain feature extraction on the frequency domain complex number sequences in the frequency domain complex number sequence set to obtain a first frequency domain feature vector, a second frequency domain feature vector, and a third frequency domain feature vector, and fusing the first frequency domain feature vector, the second frequency domain feature vector, and the third frequency domain feature vector to obtain a multidimensional frequency domain feature vector;

[0067] The time domain features of the operating data are extracted to obtain a multi-dimensional time domain feature vector, and the multi-dimensional frequency domain feature vector and the multi-dimensional time domain feature vector are spliced ​​in a preset order to obtain a multi-dimensional feature vector.

[0068] In one implementation, the time-domain signal acquisition method is as follows: Pressure signals are acquired via a pressure sensor (such as a strain gauge) mounted on the barrel or screw, typically at a sampling frequency of 100-1000 Hz. Torque signals are acquired via a torque sensor or servo motor current conversion (motor torque is proportional to current). Displacement signals are acquired via a linear displacement sensor (such as a linear encoder) or a servo motor encoder to record screw position changes.

[0069] In one implementation, the Fast Fourier Transform (FFT) algorithm is used to convert time-domain signals (pressure, torque, and displacement curves that vary over time) into frequency-domain complex sequences, separating the amplitude and phase information of different frequency components. This fusion of complementary information from the time and frequency dimensions provides dual characterization of overall trends and local frequency anomalies, addressing the one-sidedness of single-domain features and improving the diagnostic model's ability to identify complex faults.

[0070] In one implementation, features such as the dominant frequency, energy distribution, and harmonic components are extracted from each frequency domain complex sequence to form a single-signal frequency domain feature vector (e.g., pressure frequency domain features and torque frequency domain features). The frequency domain feature vectors of the three signal types are concatenated by signal type to generate a multidimensional frequency domain feature vector (e.g., a combination of frequency domain features for pressure, torque, and displacement). The energy distribution of frequency components (e.g., the proportion of high-frequency energy) is quantified to accurately locate fault-related characteristic frequencies (e.g., those for valve core wear), improving the ability to distinguish between fault types (e.g., radial and axial wear). The frequency domain information of multiple signals is integrated to reduce interference from individual signals and enhance the robustness of the features.

[0071] In one implementation, quantitative features are extracted from the time domain signals of screw pressure, torque, and displacement, for example: mean: the average value of the signal, reflecting the trend component, abnormal wear may cause mean shift (such as a decrease in the mean pressure indicates melt leakage); root mean square: a measure of signal energy, wear may cause abnormal energy distribution (such as an increase in torque RMS indicates load fluctuation); maximum: signal peak, wear may cause abnormal impact to cause the maximum value to exceed the limit (such as an abnormal increase in the maximum displacement indicates abnormal screw movement); minimum: signal valley, combined with the maximum value to reflect the fluctuation range (such as a decrease in the minimum pressure value may be due to seal failure); peak-to-peak value: maximum value - minimum value, directly reflects the signal fluctuation amplitude, the more severe the wear, the larger the peak-to-peak value is usually; standard deviation: the degree of signal dispersion, an increase in standard deviation indicates intensified fluctuation (such as an increase in torque standard deviation reflects unstable melt reflux); absolute average: the average of the absolute values, insensitive to positive and negative signal fluctuations, used to detect continuous offsets (such as an abnormal absolute average displacement indicates screw position deviation). Dimensionless features: kurtosis, skewness, waveform factor, pulse factor, and margin factor; thus reflecting the multi-dimensional time-domain feature vector of signal amplitude, fluctuation range, impact, and morphological symmetry; by quantifying the time-domain statistical laws of the signal, it can capture abnormal pressure fluctuations, torque mutations, or displacement instability caused by check valve wear in real time, significantly improving fault diagnosis accuracy; dimensionless features eliminate the influence of physical dimensions, adapt to cross-scenario analysis under different working conditions, and achieve early warning and accurate identification of wear status.

[0072] In one implementation, multidimensional time-domain feature vectors are concatenated with multidimensional frequency-domain feature vectors in a predetermined order (for example, time domain first, then frequency domain, with the same signal features arranged consecutively) to generate a composite feature vector containing both time-domain statistics and frequency-domain information. This reduces the susceptibility of single-domain features to noise or process fluctuations (e.g., temperature fluctuations leading to pressure fluctuations in the time domain). Multi-domain fusion reduces false positive rates through cross-validation.

[0073] In one embodiment, weighted fusion is performed on each vibration data to obtain one-dimensional data, including:

[0074] Get the vibration data x of each sensor i (T) and normalize the vibration data to obtain the target vibration data X i (T), generates a symmetric matrix S according to each sensor;

[0075] The correlation coefficient r between the target sensor i and each sensor is calculated based on the symmetric matrix S i (T), the average value is calculated based on the vibration data of each sensor and standard deviation σ i (T), according to the mean and standard deviation σ i (T) Calculate the variation factor;

[0076] According to the correlation coefficient r i (T) and the mutation factor are used to calculate the fusion weight ω i (T), normalize the fusion weight to get the target weight ω i '(T), according to the target weight ω i '(T) For each vibration data x i (T) Perform weighted summation to obtain one-dimensional data.

[0077] In one implementation, the vibration data x of each sensor is obtained. i (T) and normalize to get the target vibration data X i (T) can unify vibration data collected by different sensors into the same dimensional range. After normalization, the vibration data from each sensor is numerically comparable, eliminating data deviations caused by differences in sensor range and sensitivity. This lays the foundation for subsequent data processing and analysis, allowing data from different sensors to be comprehensively analyzed under the same standard, avoiding misleading subsequent calculation results due to differences in data dimensionality.

[0078] In one implementation, a symmetric matrix S is generated based on the normalized vibration data. This matrix reflects the relationships between the vibration data from each sensor. This symmetric matrix allows for a quantitative representation of the relationship between each sensor and all others, and the matrix's symmetry ensures bidirectional consistency of these relationships. This approach clearly demonstrates the correlation patterns between sensors, providing a structured data foundation for further analysis of inter-sensor correlations.

[0079] In one implementation, the correlation coefficient r between the target sensor i and each sensor is calculated based on the symmetric matrix S i (T), which can quantitatively measure the linear correlation between the vibration data of the target sensor and other sensors; high r i (T) sensor (such as r i >0.7) are given higher weights, and data with strong group consistency dominate the fusion results, reducing the impact of single sensor failure on the overall analysis.

[0080] In one implementation, the quality of sensor data is dynamically measured by the variation factor to identify abnormal or low reliability sensors. i (T) and the variation factor to calculate the fusion weight ω i (T);

[0081] Correlation coefficient r i (T): , where ri(T) represents the correlation of sensor i, m is the number of sensors, and sij represents the correlation coefficient between sensors i and j;

[0082] , where X i (T), X j (T) represents the normalized data of sensor i and j, Cov(X i (T),X j (T)) represents the covariance of sensor i and j data, D(X i (T))、D(X j (T)) represents the variance of the data of sensors i and j.

[0083] Factor of variation: , B i (T) is the variation factor of sensor i, σ i (T) is the standard deviation of sensor i in the time window, represents the mean value of sensor i in the time window;

[0084] Fusion weight ω i (T): ,ω i (T) represents the fusion weight of sensor i, ri(T) represents the correlation of sensor i, B i (T) is the variation factor of sensor i;

[0085] Fusion weights take into account both the correlation between sensors and the stability of their data. Sensors with high correlation coefficients are assigned higher weights during the fusion process to reflect their synergy with other sensors. Sensors with large variance factors are also given higher weights to reflect the stability and reliability of their data. In this way, fusion weights optimally balance sensor correlation and data stability, ensuring that the fusion results more accurately reflect the overall vibration state of the system and preventing the fusion results from being significantly affected by abnormal or unstable data from individual sensors.

[0086] In one implementation, according to the target weight ω i ′(T) for each vibration data x i (T) performs weighted summation to generate one-dimensional data. This process fuses the vibration data from multiple sensors into a single one-dimensional data set, achieving data dimensionality reduction. Through weighted summation, the fusion result can comprehensively reflect the characteristics of the vibration data from each sensor, while preserving the correlation between sensors and data stability information. The generation of one-dimensional data simplifies complex multi-sensor vibration data, facilitating subsequent analysis and processing. It can more intuitively reflect the overall vibration state of the system, providing concise and effective data support for applications such as fault diagnosis and condition monitoring, and improving data processing efficiency.

[0087] In one embodiment, reconstructing the one-dimensional data to obtain vibration characteristic data includes:

[0088] Gaussian white noise is obtained and added to one-dimensional data to obtain target one-dimensional data. The target one-dimensional data is subjected to empirical mode decomposition to obtain multiple intrinsic mode components, and the intrinsic mode components are averaged to obtain a first-order intrinsic mode function.

[0089] A first residual signal is calculated based on the preprocessed vibration data and the first-order intrinsic mode function, and an empirical mode decomposition is performed on the first residual signal to obtain a second-order intrinsic mode function;

[0090] The first residual signal is subtracted according to the second-order intrinsic mode function to obtain the second residual signal, the second residual signal is iteratively decomposed to obtain the N-order intrinsic mode function, and the N-1th residual signal is subtracted according to the N-order intrinsic mode function to obtain the Nth residual signal. When the residual is a monotonic trend term, the Nth residual signal is used as the vibration characteristic data.

[0091] In one implementation, the modal aliasing problem of traditional EMD is alleviated by adding Gaussian white noise to the one-dimensional data and generating the target one-dimensional data. The IMF components are averaged after multiple decompositions (for example: , where J represents the total number of times noise is added, IMFj(t) represents the first-order IMF component obtained by the j-th EMD decomposition, and IMF1(t) represents the first-order intrinsic mode function). This eliminates the randomness introduced by a single noise, making the generated first-order intrinsic mode function (IMF1) purer.

[0092] In one implementation, the second residual signal is iteratively decomposed to obtain an Nth order intrinsic mode function, and the N-1th residual signal is subtracted from the Nth order intrinsic mode function to obtain the Nth residual signal, for example: N (t)=r N-1 (t)-IMF N , where r N (t) is the Nth residual signal, r N-1 (t) is the N-1th residual signal, IMF N is the N-order intrinsic mode function; by subtracting the IMF component step by step, the different frequency components in the signal are gradually stripped off until the residual is a monotonic trend term.

[0093] In one implementation, the iterative residual decomposition process covers the vibration characteristics of the entire frequency band by gradually stripping off the high-frequency to low-frequency components. The first residual signal retains the mid- and low-frequency information to avoid high-frequency noise interference; the generation of high-order IMF components focuses on the impact characteristics of a specific frequency band (such as the mechanical collision of a check valve).

[0094] In one implementation, decomposition is iterated until the residual has a monotonic trend, ensuring that no signal components are missed. The final residual contains low-frequency baseline vibration (such as the background noise of the injection molding machine during steady-state operation), providing a reference for fault diagnosis. Decomposition is terminated when the residual signal becomes monotonic, avoiding computational redundancy caused by excessive decomposition.

[0095] In one embodiment, performing status monitoring on the check valve according to the fault detection result includes:

[0096] If the multi-dimensional feature vector is determined to be abnormal and the vibration feature data is determined to be normal, it is determined to be a non-check valve fault;

[0097] If the multi-dimensional feature vector is determined to be normal and the vibration feature data is determined to be abnormal, it is determined that the check valve is early worn;

[0098] If the multi-dimensional feature vector is determined to be abnormal and the vibration feature data is determined to be abnormal, it is determined that the check valve is severely worn.

[0099] In one implementation, screw anomalies other than check valve failure are often caused by process parameter fluctuations (such as uneven material temperature or insufficient back pressure) or screw / barrel wear, rather than mechanical damage to the check valve (no abnormal vibration). For example, scratches on the barrel wall can cause melt retention and pressure fluctuations, but the check valve seals properly; high material humidity can create bubbles, causing torque fluctuations, but the check valve is not worn. Check process parameters (material temperature, back pressure, and speed); check for screw wear or barrel damage, and temporarily postpone disassembly of the check valve for inspection.

[0100] In one implementation, the check valve is experiencing early wear. Mechanical wear has occurred but has not affected the melt seal. For example, microcracks or slight radial wear (clearance 0.05-0.1mm) may develop on the valve core surface, causing vibration from impacting the valve seat during opening and closing, but the leakage is insufficient to cause abnormal screw data. A decrease in spring elasticity may delay the valve core's return, generating high-frequency vibration (e.g., 150Hz) but not hindering melt flow. Shorten the monitoring cycle and record vibration trends (e.g., an increase in amplitude for three consecutive cycles is a warning sign). Schedule an inspection of the check valve during the next downtime to prevent further wear.

[0101] In one implementation, severe check valve wear can lead to loss of sealing function or structural damage, directly causing significant melt backflow and mechanical impact. For example, radial wear clearance greater than 0.3mm can cause melt backflow during injection, preventing pressure from building and causing severe screw vibration due to reaction force. Alternatively, the valve core can be broken or stuck in the open position, causing the screw to slip (torque drop) while idling and producing metallic clashing sounds (abnormal vibration). Immediately shut down the machine and disconnect power; forced operation is prohibited. Disassemble the machine to inspect the check valve for wear (e.g., measuring the valve core diameter and seat clearance). Replace worn parts and recalibrate process parameters.

[0102] Based on the same inventive concept, the present invention also provides a state monitoring system for injection molding equipment. Figure 2 , Figure 2 A schematic structural diagram of a state monitoring system for injection molding equipment provided by an embodiment of the present invention includes: an operation data acquisition module, a vibration data acquisition module, and a fault monitoring module:

[0103] An operating data acquisition module is used to acquire operating data of the injection molding equipment and convert the operating data into a multi-dimensional feature vector; the operating data includes: pressure data, torque data and displacement data;

[0104] A vibration data acquisition module is used to obtain a vibration data set of the check valve during operation, perform weighted fusion on each vibration data to obtain one-dimensional data, and reconstruct the one-dimensional data to obtain vibration characteristic data; the vibration data set contains vibration data from multiple sensors;

[0105] The fault monitoring module is used to use the multidimensional feature vector and the vibration feature data as inputs of the fault detection model to obtain a fault detection result, and perform state monitoring on the check valve according to the fault detection result.

[0106] A condition monitoring system for injection molding equipment provided in an embodiment of the present invention integrates the operating data of the injection molding machine and the vibration data of the check valve to generate a comprehensive feature data set for fault detection, thereby achieving real-time and accurate condition monitoring of the injection molding equipment, improving the reliability of equipment operation and maintenance efficiency, and reducing the failure rate and maintenance costs.

[0107] In one embodiment, the operation data acquisition module includes: a frequency domain conversion module, a feature fusion module and a feature splicing module:

[0108] A frequency domain conversion module is used to perform fast Fourier transform on the pressure data, torque data and displacement data in the operating data to obtain a frequency domain complex sequence set; the frequency domain complex sequence set includes: a first frequency domain complex sequence, a second frequency domain complex sequence and a third frequency domain complex sequence;

[0109] a feature fusion module, configured to extract frequency domain features from the frequency domain complex sequence in the frequency domain complex sequence set to obtain a first frequency domain feature vector, a second frequency domain feature vector, and a third frequency domain feature vector, and fuse the first frequency domain feature vector, the second frequency domain feature vector, and the third frequency domain feature vector to obtain a multidimensional frequency domain feature vector;

[0110] The feature splicing module is used to extract time domain features of the operating data to obtain a multidimensional time domain feature vector, and to splice the multidimensional frequency domain feature vector and the multidimensional time domain feature vector in a preset order to obtain a multidimensional feature vector.

[0111] In one embodiment, the vibration data acquisition module includes: a first execution, a second execution, and a third execution:

[0112] The first execution module is used to obtain the vibration data x of each sensor i (T) and normalize the vibration data to obtain the target vibration data X i (T), generates a symmetric matrix S according to each sensor;

[0113] The second execution module is used to calculate the correlation coefficient r between the target sensor i and each sensor according to the symmetric matrix S i (T), the average value is calculated based on the vibration data of each sensor and standard deviation σ i (T), according to the mean and standard deviation σ i (T) Calculate the variation factor;

[0114] The third execution module is used to calculate the correlation coefficient r i (T) and the mutation factor are used to calculate the fusion weight ω i (T), normalize the fusion weight to get the target weight ω i '(T), according to the target weight ω i '(T) For each vibration data x i (T) Perform weighted summation to obtain one-dimensional data.

[0115] In one embodiment, the vibration data acquisition module further includes: a modal decomposition module, a primary decomposition module, and an iterative decomposition module:

[0116] A modal decomposition module is used to obtain Gaussian white noise and add it to one-dimensional data to obtain target one-dimensional data, decompose the target one-dimensional data into multiple intrinsic modal components through empirical mode decomposition, and average each intrinsic modal component to obtain a first-order intrinsic mode function;

[0117] A primary decomposition module is used to calculate a first residual signal based on the preprocessed vibration data and the first-order intrinsic mode function, and perform empirical mode decomposition on the first residual signal to obtain a second-order intrinsic mode function;

[0118] The iterative decomposition module is used to subtract the first residual signal according to the second-order intrinsic mode function to obtain the second residual signal, iteratively decompose the second residual signal to obtain the N-order intrinsic mode function, subtract the N-1th residual signal according to the N-order intrinsic mode function to obtain the Nth residual signal, and use the Nth residual signal as the vibration characteristic data until the residual is a monotonic trend term.

[0119] In one embodiment, the fault monitoring module includes: a first judgment module, a second judgment module and a third judgment module:

[0120] A first judgment module is configured to determine that the fault is not a check valve fault if the multi-dimensional feature vector is judged to be abnormal and the vibration feature data is judged to be normal;

[0121] A second judgment module is configured to determine that the check valve is in early stage of wear if the multi-dimensional feature vector is judged to be normal and the vibration feature data is judged to be abnormal;

[0122] The third judgment module is configured to determine that the check valve is severely worn if the multi-dimensional feature vector is judged to be abnormal and the vibration feature data is judged to be abnormal.

[0123] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A method for monitoring the state of injection molding equipment, characterized in that: The method comprises: Acquiring operating data of the injection molding equipment and converting the operating data into a multi-dimensional feature vector; the operating data includes: pressure data, torque data and displacement data; Acquire a vibration data set of the check valve during operation, perform weighted fusion on each vibration data to obtain one-dimensional data, and reconstruct the one-dimensional data to obtain vibration characteristic data; the vibration data set includes vibration data of multiple sensors; Using the multidimensional feature vector and the vibration feature data as inputs of a fault detection model to obtain a fault detection result, and performing condition monitoring on the check valve according to the fault detection result; Converting the operating data into a multi-dimensional frequency domain feature vector includes: Performing fast Fourier transform on the pressure data, torque data, and displacement data in the operation data to obtain a frequency domain complex sequence set; the frequency domain complex sequence set includes: a first frequency domain complex sequence, a second frequency domain complex sequence, and a third frequency domain complex sequence; Performing frequency domain feature extraction on the frequency domain complex number sequences in the frequency domain complex number sequence set to obtain a first frequency domain feature vector, a second frequency domain feature vector, and a third frequency domain feature vector, and fusing the first frequency domain feature vector, the second frequency domain feature vector, and the third frequency domain feature vector to obtain a multidimensional frequency domain feature vector; Performing time domain feature extraction on the operating data to obtain a multidimensional time domain feature vector, and splicing the multidimensional frequency domain feature vector and the multidimensional time domain feature vector in a preset order to obtain a multidimensional feature vector; Reconstructing the one-dimensional data to obtain vibration characteristic data includes: Obtaining Gaussian white noise and adding the Gaussian white noise to the one-dimensional data to obtain target one-dimensional data, performing empirical mode decomposition on the target one-dimensional data to obtain a plurality of intrinsic mode components, and averaging the intrinsic mode components to obtain a first-order intrinsic mode function; A first residual signal is calculated based on the preprocessed vibration data and the first-order intrinsic mode function, and an empirical mode decomposition is performed on the first residual signal to obtain a second-order intrinsic mode function; The first residual signal is subtracted according to the second-order intrinsic mode function to obtain the second residual signal, the second residual signal is iteratively decomposed to obtain the N-order intrinsic mode function, and the N-1th residual signal is subtracted according to the N-order intrinsic mode function to obtain the Nth residual signal. When the residual is a monotonic trend term, the Nth residual signal is used as the vibration characteristic data.

2. A method for monitoring the state of injection molding equipment according to claim 1, characterized in that: The vibration data are weighted and fused to obtain one-dimensional data, including: Get the vibration data x of each sensor i (T) and normalize the vibration data to obtain the target vibration data X i (T), generates a symmetric matrix S according to each sensor; The correlation coefficient r between the target sensor i and each sensor is calculated based on the symmetric matrix S i (T), the average value is calculated based on the vibration data of each sensor and standard deviation σ i (T), according to the mean and standard deviation σ i (T) Calculate the variation factor; According to the correlation coefficient r i (T) and the mutation factor are used to calculate the fusion weight ω i (T), normalize the fusion weight to get the target weight ω i '(T), according to the target weight ω i '(T) For each vibration data x i (T) Perform weighted summation to obtain one-dimensional data.

3. A method for monitoring the state of injection molding equipment according to claim 1, characterized in that: Performing status monitoring on the check valve according to the fault detection result includes: If the multi-dimensional feature vector is determined to be abnormal and the vibration feature data is determined to be normal, it is determined to be a non-check valve fault; If the multidimensional feature vector is determined to be normal and the vibration feature data is determined to be abnormal, it is determined that the check valve is early worn; If the multi-dimensional feature vector is determined to be abnormal and the vibration feature data is determined to be abnormal, it is determined that the check valve is severely worn.

4. A condition monitoring system for injection molding equipment, characterized in that: The system includes: an operation data acquisition module, a vibration data acquisition module and a fault monitoring module: The operation data acquisition module is used to acquire the operation data of the injection molding equipment and convert the operation data into a multi-dimensional feature vector; the operation data includes: pressure data, torque data and displacement data; The vibration data acquisition module is used to obtain a vibration data set of the check valve during operation, perform weighted fusion on each vibration data to obtain one-dimensional data, and reconstruct the one-dimensional data to obtain vibration characteristic data; the vibration data set includes vibration data of multiple sensors; The fault monitoring module is configured to use the multidimensional feature vector and the vibration feature data as inputs of a fault detection model to obtain a fault detection result, and perform state monitoring on the check valve according to the fault detection result; The operation data acquisition module includes: a frequency domain conversion module, a feature fusion module and a feature splicing module: The frequency domain conversion module is configured to perform fast Fourier transform on the pressure data, torque data, and displacement data in the operation data to obtain a frequency domain complex sequence set; the frequency domain complex sequence set includes: a first frequency domain complex sequence, a second frequency domain complex sequence, and a third frequency domain complex sequence; The feature fusion module is used to perform frequency domain feature extraction on the frequency domain complex number sequences in the frequency domain complex number sequence set to obtain a first frequency domain feature vector, a second frequency domain feature vector and a third frequency domain feature vector, and fuse the first frequency domain feature vector, the second frequency domain feature vector and the third frequency domain feature vector to obtain a multidimensional frequency domain feature vector; The feature splicing module is used to extract time domain features from the operating data to obtain a multidimensional time domain feature vector, and to splice the multidimensional frequency domain feature vector and the multidimensional time domain feature vector in a preset order to obtain a multidimensional feature vector; The vibration data acquisition module further includes: a modal decomposition module, a primary decomposition module and an iterative decomposition module: The modal decomposition module is used to obtain Gaussian white noise and add the Gaussian white noise to the one-dimensional data to obtain target one-dimensional data, obtain multiple intrinsic modal components by empirical mode decomposition of the target one-dimensional data, and average the intrinsic modal components to obtain a first-order intrinsic mode function; The primary decomposition module is used to calculate a first residual signal based on the preprocessed vibration data and the first-order intrinsic mode function, and perform empirical mode decomposition on the first residual signal to obtain a second-order intrinsic mode function; The iterative decomposition module is used to subtract the first residual signal according to the second-order intrinsic mode function to obtain the second residual signal, iteratively decompose the second residual signal to obtain the N-order intrinsic mode function, subtract the N-1th residual signal according to the N-order intrinsic mode function to obtain the Nth residual signal, until the residual is a monotonic trend term, then the Nth residual signal is used as the vibration characteristic data.

5. A condition monitoring system for injection molding equipment according to claim 4, characterized in that: The vibration data acquisition module includes: a first execution, a second execution and a third execution: The first execution module is used to obtain the vibration data x of each sensor i (T) and normalize the vibration data to obtain the target vibration data X i (T), generates a symmetric matrix S according to each sensor; The second execution module is used to calculate the correlation coefficient r between the target sensor i and each sensor according to the symmetric matrix S i (T), the average value is calculated based on the vibration data of each sensor and standard deviation σ i (T), according to the mean and standard deviation σ i (T) Calculate the variation factor; The third execution module is used to calculate the correlation coefficient r i (T) and the mutation factor are used to calculate the fusion weight ω i (T), normalize the fusion weight to get the target weight ω i '(T), according to the target weight ω i '(T) For each vibration data x i (T) Perform weighted summation to obtain one-dimensional data.

6. A condition monitoring system for injection molding equipment according to claim 4, characterized in that: The fault monitoring module includes: a first judgment module, a second judgment module and a third judgment module: The first judgment module is configured to determine that the fault is not a check valve fault if the multidimensional feature vector is judged to be abnormal and the vibration feature data is judged to be normal; The second judgment module is configured to determine that the check valve is in early stage of wear if the multi-dimensional feature vector is judged to be normal and the vibration feature data is judged to be abnormal; The third judgment module is configured to determine that the check valve is severely worn if the multi-dimensional feature vector is judged to be abnormal and the vibration feature data is judged to be abnormal.

Images