Real-time monitoring system for cutting state of gantry machining center based on multi-sensor fusion
By constructing a directed graph structure in the gantry machining center and selecting the optimal dependent variable to establish a multi-sensor correlation model, the problem of insufficient monitoring accuracy was solved, and real-time and accurate monitoring of the cutting state was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JINAN ZHANGLI MACHINERY
- Filing Date
- 2026-03-31
- Publication Date
- 2026-06-09
AI Technical Summary
Existing multi-sensor fusion-based methods for monitoring the cutting status of gantry machining centers suffer from insufficient monitoring accuracy. In particular, the physical relationship between dependent and independent variables is affected by dynamic factors under different cutting stages or working conditions, leading to increased model prediction errors.
By deploying sensors at different locations in a gantry machining center, extracting feature quantities and constructing a directed graph structure, determining the optimality of nodes based on the spatial location and response relationship of the sensors, selecting the optimal dependent variable to establish a multi-sensor association model, and realizing real-time monitoring of the cutting state.
It significantly improves the accuracy of cutting condition monitoring, avoids the failure of dependent variables under different cutting stages or working conditions, and ensures that the model dependent variables can fully and evenly reflect the combined effect of multi-sensor signals.
Smart Images
Figure CN122165236A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of machine tool monitoring data processing technology, and in particular to a real-time monitoring system for the cutting status of a gantry machining center based on multi-sensor fusion. Background Technology
[0002] The cutting process is a core element affecting the production efficiency and machining quality of gantry machining centers. Therefore, the ability to effectively monitor the cutting state directly impacts the machine tool's operational safety and production efficiency. During the cutting process, when the tool contacts the workpiece and material is removed, cutting force is generated, accompanied by energy transfer and release. This cutting action is a multi-physics coupling process, and the signals from various sensors are inherently correlated under normal cutting conditions. Therefore, in traditional methods, multi-sensor data is first collected within a preset time window, and statistical features that characterize signal strength, such as root mean square values or energy feature values, are extracted. Then, a specific sensor feature is fixedly selected as the dependent variable, and a multiple regression or similar correlation model is established using the remaining sensor features. The residual between the model output and the actual measured value of the selected dependent variable is used to achieve real-time monitoring of the cutting state.
[0003] However, in actual gantry machining center cutting processes, the interaction between the tool and the workpiece is highly dynamic and non-uniform. Differences in the process, the feed rate, and the different machining stages all directly change the force distribution on the tool and the cutting energy transfer path. For example, in the roughing stage, due to the large depth of cut and high feed rate, the cutting force increases significantly. At this time, the spindle current signal usually responds stably to the cutting state and is easy to model. However, the machine tool structure vibration and tool acoustic emission signal may be affected by cutting impact, local friction and structural resonance, exhibiting transient peaks and large fluctuations. In the finishing stage, the cutting force decreases and the amount of material removed decreases, but tool wear gradually accumulates, making the vibration and acoustic emission signals more sensitive to abnormal cutting conditions. At this time, some sensor signals may experience local interference, sudden vibration or reduced signal-to-noise ratio, which weakens the physical correlation between the originally fixed dependent variable and the other sensor characteristic quantities. In addition, the dynamic characteristics of the machine tool structure (such as spindle rigidity, slide rail guiding characteristics, bed vibration frequency response) may also cause nonlinear changes in signal transmission under different cutting conditions. Changes in cutting parameters (such as cutting speed, feed rate, and depth of cut) will also cause the amplitude, frequency and phase of each sensor to drift for the same cutting event.
[0004] Therefore, in the complex environment of the above-mentioned factors, the traditional multi-sensor fusion monitoring method with fixed dependent variables may still have an increased model prediction error, even under normal cutting conditions, because the physical relationship between the dependent and independent variables may change due to the influence of the above-mentioned dynamic factors. This results in a large residual and affects the accuracy and stability of the cutting condition judgment.
[0005] In other words, the current method for monitoring the cutting status of gantry machining centers based on multi-sensor fusion suffers from insufficient monitoring accuracy. Summary of the Invention
[0006] In view of this, the present invention provides a real-time monitoring system for cutting status of gantry machining centers based on multi-sensor fusion, in order to solve the technical problem of insufficient monitoring accuracy in current multi-sensor fusion-based methods for monitoring cutting status of gantry machining centers.
[0007] The present invention provides a real-time monitoring system for cutting status of a gantry machining center based on multi-sensor fusion, comprising:
[0008] The sensor data feature quantity determination module is used to deploy different types of sensors at different locations in the gantry machining center, select a preset time period before the current moment as a time window, and calculate the pre-determined feature quantity corresponding to the type of each sensor data within the time window.
[0009] The directed graph structure determination module is used to take the feature quantities corresponding to each sensor under the time window as nodes and determine the directed graph structure of each node in combination with the spatial position of the sensor.
[0010] The node preference determination module is used to determine the reliability of the response result of any node based on the directed graph structure of any node, according to the number and weight of the incoming edges and the number of outgoing edges of any node, to determine the balance of the incoming edge weight distribution of any node according to the value range and overall value deviation of the incoming edge weight of any node, and to determine the preference of any node according to the reliability of the response result and the balance of the incoming edge weight distribution.
[0011] The monitoring result acquisition module is used to select the node with the highest optimization degree as the optimal dependent variable, and to complete the monitoring of the cutting state of the gantry machining center based on the optimal dependent variable.
[0012] Furthermore, the feature quantities include root mean square value and mean value, and the type of feature quantity corresponding to each type of sensor data is unique.
[0013] Furthermore, the determination of the directed graph structure of each node based on the spatial location of the sensors includes:
[0014] The node vector corresponding to any node is constructed by taking the feature value corresponding to any node and the three-dimensional spatial coordinates of the sensor corresponding to any node as vector dimensions.
[0015] Based on the node vectors corresponding to all nodes, a directed graph structure for each node is generated using the partial least squares regression method.
[0016] Furthermore, determining the reliability of the response result of any of the nodes includes:
[0017] Determine the proportion of the number of incoming edges of any node to the total number of edges connected to any node and record it as the first proportion. Determine the proportion of the number of outgoing edges of any node to the total number of edges connected to any node and record it as the second proportion. An incoming edge is an edge that points from other nodes to any node, and an outgoing edge is an edge that points from any node to other nodes.
[0018] Calculate the sum of all incoming edge weights of any node, and determine a first confidence characterization value based on the sum of all incoming edge weights of any node and the first proportion. The first confidence characterization value is proportional to both the sum of all incoming edge weights of any node and the first proportion.
[0019] A second credibility characterization value is determined based on the second proportion, and the second credibility characterization value is inversely proportional to the second proportion;
[0020] The credibility of the response result of any node is determined based on the first credibility characterization value and the second credibility characterization value, wherein the credibility of the response result is proportional to both the first credibility characterization value and the second credibility characterization value.
[0021] Furthermore, determining the balance of the incoming edge weight distribution of any node includes:
[0022] The difference between the maximum and minimum values of the incoming edge weights of any node is recorded as the first balance determination value. The absolute value of the difference between any incoming edge weight of any node and the mean of all incoming edge weights of any node is calculated and recorded as the deviation of the value of any incoming edge weight. The sum of the deviations of the values of all incoming edge weights of any node is recorded as the second balance determination value.
[0023] The balance of the incoming edge weight distribution of any node is determined based on the first balance determination value and the second balance determination value, wherein the balance of the incoming edge weight distribution is inversely proportional to both the first balance determination value and the second balance determination value.
[0024] Furthermore, determining the preference of any given node includes:
[0025] The optimality of any node is directly proportional to the reliability of its response and the balance of its incoming edge weight distribution.
[0026] Furthermore, based on the optimal dependent variable, the cutting state of the gantry machining center is monitored, including:
[0027] Nodes other than the optimal dependent variable are designated as other nodes. A correlation model is established for each of the other nodes with respect to the optimal dependent variable. The residual between the output of the correlation model and the actual value of the optimal dependent variable is used to monitor the cutting status of the gantry machining center.
[0028] The advantages of this invention compared to the prior art are:
[0029] This invention extracts feature quantities from sensor data deployed at different locations in a gantry machining center within a time window. Then, using each sensor's feature quantity as a node, and combining this with the sensor's spatial location, a directed graph structure is determined for each node. Based on the proportion and weight of each node's incoming edges among all edges, and the deviation of the weight distribution, the optimality of each node as the dependent variable in the multi-sensor correlation model is determined. The node with the highest optimality is used as the optimal dependent variable to complete the cutting condition monitoring. This invention ensures that the dependent variable of the established multi-sensor correlation model can fully and evenly reflect the combined effect of multiple sensor signals, avoiding the failure of fixed dependent variables under different cutting stages or working conditions, and significantly improving the accuracy of cutting condition monitoring in gantry machining centers. Attached Figure Description
[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0031] Figure 1 This is a structural block diagram of a real-time monitoring system for cutting status of a gantry machining center based on multi-sensor fusion, provided in Embodiment 1 of the present invention. Detailed Implementation
[0032] The overall concept of this invention is as follows:
[0033] This invention proposes a real-time monitoring system for the cutting status of a gantry machining center based on multi-sensor fusion. It quantifies the relative spatial positions of different sensors within the gantry machining center by establishing a spatial coordinate system. Based on this coordinate system, a directed graph structure is constructed between the sensors to provide a basis for the rational selection of the state representation benchmark in the subsequent cutting status monitoring model. Then, according to the constructed directed graph structure, the causal response index and the balance of the incoming edge weight distribution of each node are obtained. Combining these two factors yields the node dependent variable suitability index. Based on the suitability index, the optimal dependent variable for different time periods is adaptively obtained. A multi-sensor correlation model is established based on the optimal dependent variable to achieve real-time monitoring of the cutting status of the gantry machining center. This ensures that the established model's dependent variable can fully and evenly reflect the combined effect of multiple sensor signals, avoiding the failure defects of fixed dependent variables under different cutting stages or working conditions, and significantly improving the accuracy of cutting status monitoring.
[0034] To further illustrate the technical solution of the present invention, specific embodiments are described below.
[0035] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of the invention include a particular feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. Furthermore, a particular feature, structure, or characteristic in one or more embodiments may be combined in any suitable form, and the terms "comprising," "including," "having," and variations thereof mean "including, but not limited to," unless otherwise specifically emphasized.
[0036] It should be understood that the sequence number of each step in the following embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.
[0037] Method Implementation Examples:
[0038] See Figure 1 This is a structural block diagram of a real-time monitoring system for the cutting status of a gantry machining center based on multi-sensor fusion, provided in Embodiment 1 of the present invention. Figure 1 As shown, the system includes a sensor data feature determination module 11, a directed graph structure determination module 12, a node optimization degree determination module 13, and a monitoring result acquisition module 14. The following is a detailed description of each module:
[0039] The sensor data feature quantity determination module 11 is used to deploy different types of sensors at different locations in the gantry machining center, select a preset time period before the current moment as a time window, and calculate the pre-determined feature quantity corresponding to the type of each sensor data within the time window.
[0040] First, in order to realize the monitoring of the cutting state of the gantry machining center based on multi-sensor fusion, it is necessary to deploy various types of sensors at different locations of the gantry machining center, such as cutting force sensors, machine tool vibration sensors, spindle current / voltage sensors, etc., to collect relevant data in real time. The specific locations of each type of sensor can be set by those skilled in the art based on the data types collected by the sensors and the performance types of relevant signals about the cutting state at various locations of the gantry machining center. This embodiment will not elaborate further.
[0041] After acquiring sensor data, to ensure the accuracy of subsequent analysis, preprocessing methods for the sensor data need to be set. For example, high-frequency signals are low-pass / band-pass filtered to remove mechanical background noise; current signals are smoothed and DC components are removed; since different sensors may have different sampling frequencies, time alignment or resampling is also required to ensure that the signals from different sensors correspond within the same time window. Furthermore, extreme outliers are removed using statistical methods, such as the 3σ principle, and imputed using linear interpolation; missing data points are also detected and imputed using linear interpolation and other interpolation methods.
[0042] After completing the sensor setup and preprocessing of the sensor data, in the subsequent real-time monitoring process, a preset time window is first used to calculate the characteristic quantity corresponding to the type of each sensor data within this time window. The preset time window duration can be set comprehensively based on monitoring accuracy and computational efficiency. Furthermore, the type of characteristic quantity to be extracted needs to be determined according to the sensor type and requirements, and each type of sensor has a unique characteristic quantity type extracted; that is, each sensor extracts only one characteristic quantity in the current time window. In this embodiment, the characteristic quantity corresponding to the vibration sensor is the root mean square (RMS), the characteristic quantity corresponding to the spindle current / voltage sensor is the mean value, and the characteristic quantity corresponding to the cutting force sensor is the mean value.
[0043] Based on the pre-determined characteristic quantities corresponding to the type of each sensor data within the calculated time window, an analysis premise is provided for subsequent real-time monitoring.
[0044] The directed graph structure determination module 12 is used to take the feature quantities corresponding to each sensor under the time window as nodes and determine the directed graph structure of each node in combination with the spatial position of the sensor.
[0045] The signals collected by various sensors are essentially the result of multi-physical responses after the cutting action between the tool and the workpiece propagates through the machine tool structure and power system. The generation and transmission of this response have clear spatial path characteristics. That is, the cutting action first occurs in the contact area between the tool and the workpiece, and then is transmitted step by step through the spindle, tool clamping structure, slide, crossbeam and bed, etc., and causes responses such as force, vibration, current or acoustic emission at different structural positions. In other words, the correlation between the signals measured by different sensors is not formed randomly, but is gradually formed by the coupling during the propagation of the cutting action along the spatial path of the machine tool structure. The degree of correlation is closely related to the spatial position of the sensor relative to the cutting action point, the length of the structural transmission path and the dynamic characteristics of the structure.
[0046] Therefore, to accurately reflect the true correlation between the characteristic quantities of each sensor and identify the sensor characteristic quantity that is more suitable as the state characterization benchmark under the current cutting state, it is necessary to start from the spatial propagation law of cutting action and the correlation structure between each sensor, quantify the relative spatial position relationship of different sensors in the gantry machining center by establishing a spatial coordinate system, and construct a directed graph structure between sensors based on this spatial coordinate system to provide a basis for the reasonable selection of the state characterization benchmark in the subsequent cutting state monitoring model. Specifically:
[0047] First, since the contact area between the tool and the workpiece is the actual location where material removal and energy action occur, and also the starting point for the generation of cutting force and the propagation of energy to the machine tool structure, the physical effects such as forces, vibrations, and stress waves generated during the cutting process first act on the tool and spindle system, and then propagate outward step by step through the spindle, ram, crossbeam, and bed, causing corresponding vibrations, current changes, and responses at different structural locations of the machine tool during the propagation process. Therefore, in this embodiment, it is preferable to establish a three-dimensional spatial coordinate system for the gantry machining center using the cutting point of the machining center tool or the actual location where cutting occurs as the origin, and record the spatial coordinates of the sensors deployed on the machining center in the current spatial coordinate system. Where x, y, and z represent the coordinates of the current sensor in each dimension of the established spatial coordinate system, respectively. Indicate the sensor type, such as vibration, current, cutting force, etc., to distinguish the spatial coordinates of different sensors.
[0048] Then, the feature quantity corresponding to each sensor in the current time window is taken as a node. That is, the feature quantity corresponding to each sensor in the current time window is taken as a node, and for any node, a node vector with spatial attributes is obtained by combining the spatial coordinates of the sensor corresponding to that node. Based on the node vector corresponding to each constructed node, a directed graph structure for each node can be generated using the Partial Least Squares Regression (PLS) method. Not all of the obtained directed graph structures necessarily contain edge connections; the direction of the edge indicates the predicted relationship between two nodes. For example, i→j means that the features of node i can be used to predict the features of node j.
[0049] Among them, constructing a directed graph structure using partial least squares regression is an existing technology. To facilitate understanding, its general process is explained here: PLS uses each sensor feature as the dependent variable in turn and the remaining features as independent variables to establish a PLS regression model. The connection relationship and direction between nodes are determined according to the degree of regression contribution of each independent variable to the dependent variable, thereby constructing a directed graph structure that reflects the predictive relationship between each sensor feature.
[0050] The node preference determination module 13 is used to determine the reliability of the response result of any node based on the directed graph structure of any node, according to the number of incoming edges and the weight of incoming edges and the number of outgoing edges of any node, to determine the balance of the incoming edge weight distribution of any node according to the value range of the incoming edge weight and the overall value deviation, and to determine the preference of any node according to the reliability of the response result and the balance of the incoming edge weight distribution.
[0051] When using a multi-sensor correlation model to monitor the cutting state of a gantry machining center, the dependent variable of the model should essentially be able to stably reflect changes in the cutting state, and its variation pattern should be effectively explained by the characteristic quantities of other related sensors. Therefore, from the perspective of correlation model construction, an ideal dependent variable should have a clear "response attribute," meaning its signal change is mainly driven by changes in other sensor signals, and its change can be reflected by establishing a stable mapping relationship through the characteristic quantities of other sensors. If a sensor signal in the response relationship primarily affects other signals, or its change is highly independent, then this signal is difficult to stably model and map through the characteristic quantities of other sensors and is unsuitable as the model's response variable; conversely, this sensor signal is more suitable as a dependent variable so that it can be better explained by other sensor signals as independent variables. In summary, the directed graph structure constructed in the previous module can be used to analyze the strength of other nodes' explanatory power and the strength of its own explanation for each node, thereby determining the reliability of each node's response results and providing a basis for determining its optimal selection as the optimal dependent variable.
[0052] However, relying solely on the reliability of node response results still has certain limitations. This is because the reliability of node response results can only reflect the overall response intensity of the node in the causal network, but it fails to reveal the structural characteristics and source distribution of the node response. Due to the dynamic changes in machining stage, cutting depth, feed rate, and tool wear state, the same node may be affected by multiple sensor signals simultaneously. If its response is mainly dominated by a very small number of nodes, the model may become overly dependent on individual inputs and fail to stably reflect the overall cutting state of the system. Therefore, it is necessary to further analyze the distribution balance of the incoming edge weights on each node, that is, to measure whether the causal effects on the node are evenly distributed among the input nodes, rather than being dominated by only a few signals.
[0053] Considering the above two aspects, the reliability of the response result of any node can first be determined based on the directed graph structure of any node, according to the number and weight of the incoming edges and the number of outgoing edges of any node, including:
[0054] Determine the proportion of the number of incoming edges of any node to the total number of edges connected to any node and record it as the first proportion. Determine the proportion of the number of outgoing edges of any node to the total number of edges connected to any node and record it as the second proportion. An incoming edge is an edge that points from other nodes to any node, and an outgoing edge is an edge that points from any node to other nodes.
[0055] Calculate the sum of all incoming edge weights of any node, and determine a first confidence characterization value based on the sum of all incoming edge weights of any node and the first proportion. The first confidence characterization value is proportional to both the sum of all incoming edge weights of any node and the first proportion.
[0056] A second credibility characterization value is determined based on the second proportion, and the second credibility characterization value is inversely proportional to the second proportion;
[0057] The credibility of the response result of any node is determined based on the first credibility characterization value and the second credibility characterization value, wherein the credibility of the response result is proportional to both the first credibility characterization value and the second credibility characterization value.
[0058] Furthermore, as a preferred embodiment, the reliability of the response result of any node is:
[0059]
[0060] in, This represents the confidence level of the response result of the j-th node (i.e., the j-th sensor), used to characterize the strength of the current node's ability to be predicted by other nodes and its interpretability. Let represent the hyperbolic tangent function, which serves as a normalization function here. Based on the range of values of the input variables of this function, its output is ultimately restricted to [0,1]. This ensures that the larger the final output, the more other nodes point to the current node, the more information the node can obtain from the signals of more other nodes, and its changes can be better interpreted by other sensors. Furthermore, the stronger the predictive ability of these nodes for the node, the easier it is to model its numerical changes. At the same time, the node is mainly driven by other variables, rather than driving other variables. A strength index representing the ability of a node to be predicted by other nodes, where This represents the ratio of the number of incoming edges to the total number of edges connected to the current node, also known as the first proportion mentioned above. This represents the number of incoming edges to the j-th node. This represents the total number of edges related to the current node, i.e., the number of edges connected to the current node. A larger first proportion indicates that the node contributes more to the system response, and its changes can be effectively interpreted by other sensor signals. Based on this, the sum of the weights of all incoming edges to the current node is then obtained. (The weights of the edges connected to any node can be directly determined based on the structure of the directed graph.) Here, the weight of the i-th incoming edge at the j-th node is represented by the sum of the incoming edge weights. This is used to represent the strength of the signal at the current node that can be interpreted by other sensor nodes. The larger the sum, the stronger the correlation between the signal change at this node and other sensor signals, and the more it resembles the response result node of the entire cutting system. This represents the number of outgoing edges from the j-th node. This represents the ratio of the number of outgoing edges of the current node to the total number of edges it connects to, also known as the second proportion mentioned above. Since the dependent variable of the model should be the object being explained, that is, it is mainly driven by other variables rather than driving other variables, this term imposes certain restrictions on the first term to ensure that the current node is mainly explained rather than explaining other nodes.
[0061] Similarly, based on the directed graph structure of any node, the balance of the incoming edge weight distribution of any node can be determined according to the range of values of the incoming edge weights of that node and the overall deviation of the values, including:
[0062] The difference between the maximum and minimum values of the incoming edge weights of any node is recorded as the first balance determination value. The absolute value of the difference between any incoming edge weight of any node and the mean of all incoming edge weights of any node is calculated and recorded as the deviation of the value of any incoming edge weight. The sum of the deviations of the values of all incoming edge weights of any node is recorded as the second balance determination value.
[0063] The balance of the incoming edge weight distribution of any node is determined based on the first balance determination value and the second balance determination value, wherein the balance of the incoming edge weight distribution is inversely proportional to both the first balance determination value and the second balance determination value.
[0064] Furthermore, as a preferred embodiment, the balance of the incoming edge weight distribution of any node is:
[0065]
[0066] in, This represents the balance of the incoming edge weight distribution of the j-th node, reflecting the balance of the causal influence of other nodes on the current node. This represents an exponential function with the natural number e as the base. Here, an inverse proportional form is used to limit the output to [0,1]. The larger the final output, the more it indicates that the node is not dominated by a few individual input nodes, but is the result of the combined action of multiple input nodes. This means that the sensor can simultaneously reflect the comprehensive state of multiple factors such as cutting force, tool wear, and machine tool structure vibration. This represents the maximum value of the incoming edge weight of the j-th node. Let represent the minimum weight of the incoming edge to the j-th node. This represents the range of values for the incoming edge weights of the node, which is also the first balance determination value mentioned above. The range of values for the incoming edge weights reflects the maximum difference in the influence of different input nodes on the node. The smaller the range of values for the incoming edge weights, the more evenly the current node is affected by all input nodes, and no single node has an excessively dominant effect on the output, resulting in higher balance. This represents the number of incoming edges to the j-th node. This represents the weight of the i-th incoming edge of the j-th node. This represents the overall deviation of the incoming edge weights of the j-th node, which is also the aforementioned second equilibrium determination value. This represents the mean weight of the incoming edges to that node. This represents the deviation of the weight of the i-th incoming edge of the j-th node from the mean. Combining these values yields the overall deviation of each incoming edge weight. A greater degree of balance in the distribution of incoming edge weights indicates that the sensor's characteristic quantity is more evenly influenced by multiple sensor signals, without any single signal dominating. As a dependent variable, it can more stably and comprehensively reflect the cutting state.
[0067] Based on the reliability of the obtained response results and the balance of the incoming edge weight distribution, the optimality of any node can be determined, including:
[0068] The optimality of any node is directly proportional to the reliability of its response and the balance of its incoming edge weight distribution.
[0069] Furthermore, as a preferred option, the preference degree of any given node is:
[0070]
[0071] in, This represents the preference degree of the j-th node.
[0072] The monitoring result acquisition module 14 is used to select the node with the highest optimization degree as the optimal dependent variable, and to complete the monitoring of the cutting state of the gantry machining center based on the optimal dependent variable.
[0073] Obtain the optimization degree of all nodes, and then use the node corresponding to the highest optimization degree as the optimal dependent variable. Based on the optimal dependent variable, complete the monitoring of the cutting state of the gantry machining center, including:
[0074] Nodes other than the optimal dependent variable are designated as other nodes. A correlation model is established for each of the other nodes with respect to the optimal dependent variable. The residual between the output of the correlation model and the actual value of the optimal dependent variable is used to monitor the cutting status of the gantry machining center.
[0075] The embodiments of the present invention construct a multi-sensor correlation model based on the obtained optimal dependent variable, which can ensure that the dependent variable of the constructed model can fully and evenly reflect the comprehensive effect of the multi-sensor signals, avoid the failure of the fixed dependent variable under different cutting stages or working conditions, and significantly improve the accuracy of cutting status monitoring of gantry machining centers.
[0076] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.
Claims
1. A real-time monitoring system for cutting status of a gantry machining center based on multi-sensor fusion, characterized in that, include: The sensor data feature quantity determination module is used to deploy different types of sensors at different locations in the gantry machining center, select a preset time period before the current moment as a time window, and calculate the pre-determined feature quantity corresponding to the type of each sensor data within the time window. The directed graph structure determination module is used to take the feature quantities corresponding to each sensor under the time window as nodes and determine the directed graph structure of each node in combination with the spatial position of the sensor. The node preference determination module is used to determine the reliability of the response result of any node based on the directed graph structure of any node, according to the number and weight of the incoming edges and the number of outgoing edges of any node, to determine the balance of the incoming edge weight distribution of any node according to the value range and overall value deviation of the incoming edge weight of any node, and to determine the preference of any node according to the reliability of the response result and the balance of the incoming edge weight distribution. The monitoring result acquisition module is used to select the node with the highest optimization degree as the optimal dependent variable, and to complete the monitoring of the cutting state of the gantry machining center based on the optimal dependent variable.
2. The real-time monitoring system for cutting status of a gantry machining center based on multi-sensor fusion as described in claim 1, characterized in that, The feature quantities include the root mean square value and the mean value, and the type of feature quantity corresponding to each type of sensor data is unique.
3. The real-time monitoring system for cutting status of a gantry machining center based on multi-sensor fusion as described in claim 1, characterized in that, The process of determining the directed graph structure of each node by combining the spatial location of the sensors includes: The node vector corresponding to any node is constructed by taking the feature value corresponding to any node and the three-dimensional spatial coordinates of the sensor corresponding to any node as vector dimensions. Based on the node vectors corresponding to all nodes, a directed graph structure for each node is generated using the partial least squares regression method.
4. The real-time monitoring system for cutting status of a gantry machining center based on multi-sensor fusion according to claim 1, characterized in that, Determining the reliability of the response result of any of the nodes includes: Determine the proportion of the number of incoming edges of any node to the total number of edges connected to any node and record it as the first proportion. Determine the proportion of the number of outgoing edges of any node to the total number of edges connected to any node and record it as the second proportion. An incoming edge is an edge that points from other nodes to any node, and an outgoing edge is an edge that points from any node to other nodes. Calculate the sum of all incoming edge weights of any node, and determine a first confidence characterization value based on the sum of all incoming edge weights of any node and the first proportion. The first confidence characterization value is proportional to both the sum of all incoming edge weights of any node and the first proportion. A second credibility characterization value is determined based on the second proportion, and the second credibility characterization value is inversely proportional to the second proportion; The credibility of the response result of any node is determined based on the first credibility characterization value and the second credibility characterization value, wherein the credibility of the response result is proportional to both the first credibility characterization value and the second credibility characterization value.
5. The real-time monitoring system for cutting status of a gantry machining center based on multi-sensor fusion according to claim 1, characterized in that, Determining the balance of the incoming edge weight distribution of any node includes: The difference between the maximum and minimum values of the incoming edge weights of any node is recorded as the first balance determination value. The absolute value of the difference between any incoming edge weight of any node and the mean of all incoming edge weights of any node is calculated and recorded as the deviation of the value of any incoming edge weight. The sum of the deviations of the values of all incoming edge weights of any node is recorded as the second balance determination value. The balance of the incoming edge weight distribution of any node is determined based on the first balance determination value and the second balance determination value, wherein the balance of the incoming edge weight distribution is inversely proportional to both the first balance determination value and the second balance determination value.
6. The real-time monitoring system for cutting status of a gantry machining center based on multi-sensor fusion according to claim 1, characterized in that, Determining the preference of any of the nodes includes: The optimality of any node is directly proportional to the reliability of its response and the balance of its incoming edge weight distribution.
7. The real-time monitoring system for cutting status of a gantry machining center based on multi-sensor fusion according to any one of claims 1 to 6, characterized in that, Based on the optimal dependent variable, the cutting state of the gantry machining center is monitored, including: Nodes other than the optimal dependent variable are designated as other nodes. A correlation model is established for each of the other nodes with respect to the optimal dependent variable. The residual between the output of the correlation model and the actual value of the optimal dependent variable is used to monitor the cutting status of the gantry machining center.