A dynamic configuration method for equipment process flow

Abstract

The present invention relates to a method for dynamically configuring an equipment process flow, which solves the technical problems of long data transmission time and low efficiency. The method adopts the following methods: a sensor collects real-time state parameters of the process flow; an edge node receives the real-time state parameters of the process flow as data to be processed; the edge node extracts features of the data to be preprocessed, matches similar data in a historical transmission database of the edge node, and uses the similar data as reference data; uses the reference data as a source and the data to be preprocessed as a target to calculate a transformation compensation sequence; defines the data to be transmitted by using similar data labels and the transformation compensation sequence; compares the sizes of the data to be preprocessed and the data to be transmitted, and finds that the data to be processed is smaller than the data to be transmitted; the edge node compresses and encodes the data to be transmitted and then sends it to a process flow configuration unit; the process flow configuration unit performs prediction analysis based on the received data to complete the process configuration. The method solves the problem well and can be used in process flow configuration.

Description

Technical Field

[0001] The present invention relates to the field of process flow configuration, and in particular to a method for dynamically configuring an equipment process flow. Background Art

[0002] Based on the current electronic manufacturing production line resource configuration, research on manufacturing unit and system load balancing, scientific management and use of production equipment; research on multi-objective optimization algorithms based on SPT, MWKR, EDD, and SST, optimize various SMT workshop production plans, form an optimized workshop production scheduling plan, and maximize the equipment's production and processing capabilities and production efficiency; research on dynamic process configuration methods under multiple optimization target conditions, adaptively adjust according to the dynamic changes of current production resources, form production line processes, equipment and plans, and monitor, track and record the execution status in real time to reduce manual intervention and reduce production costs.

[0003] The existing equipment process flow dynamic configuration method has the technical problems of long data transmission time and low efficiency. The present invention provides an equipment process flow dynamic configuration method to solve this problem. Summary of the Invention

[0004] The technical problem to be solved by the present invention is the long data transmission time and low efficiency in the prior art. A new equipment process flow dynamic configuration method is provided, which has the characteristics of long data transmission time and low efficiency.

[0005] In order to solve the above technical problems, the technical solutions adopted are as follows:

[0006] A method for dynamically configuring an equipment process flow, the method comprising:

[0007] Step 1: The sensor collects the real-time status parameters of the process flow;

[0008] Step 2: The edge node receives the real-time status parameters of the process flow, which are defined as data to be processed;

[0009] Step 3: The edge node extracts the features of the data to be preprocessed, searches for similar data in the edge node's historical transmission database, records the similar data labels, and uses the similar data as reference data. If no similar data can be retrieved, the data to be preprocessed is defined as the data to be transmitted, and step 6 is directly executed.

[0010] Step 4: Using the reference data as the source and the data to be preprocessed as the target, calculate the transformation compensation sequence;

[0011] Step 5: Define the data to be transmitted by using similar data labels and transformation compensation sequences; compare the sizes of the data to be preprocessed and the data to be transmitted. If the data to be processed is smaller than the data to be transmitted, then execute step 6; otherwise, update the data to be preprocessed to the data to be transmitted.

[0012] Step 6: Determine the size of the data to be transmitted and the transmission threshold. If the data to be transmitted is smaller than the transmission threshold, return to step 2; otherwise, go to step 7.

[0013] Step 7: The edge node compresses and encodes the data to be transmitted and outputs it to the process flow configuration unit;

[0014] Step 8: The process flow configuration unit performs prediction analysis based on the received data and completes the process flow configuration.

[0015] In order to scientifically manage and use production equipment, form more optimized workshop scheduling plans, and dynamically and adaptively adjust according to current production resources to maximize the equipment's production and processing capabilities and production efficiency, the present invention intends to conduct research from the aspects of load balance and dynamic balance of manufacturing units.

[0016] In the above scheme, for optimization, step 2 further includes:

[0017] Step 2.1, define a feature extraction box with a width of w;

[0018] Step 2.2, use the feature extraction box to traverse the data to be preprocessed and complete feature extraction;

[0019] In step 2.3, the extracted features are matched with the historical data features in the edge node historical transmission database, and the historical data with a matching rate greater than a preset threshold is defined as similar data.

[0020] Further, step 3 includes:

[0021] Step 3.1, determining the source data sub-element in the source data and the target data sub-element in the target data using the feature extraction frame as a unit;

[0022] Step 3.2, define the transformation equation as: E = d + η × s; the compensation equation is: I'' = αI' + β;

[0023] Wherein, η is the preset weight value, I' represents the source data element after deformation, and I'' represents the source data element after amplitude compensation; the connectivity s of adjacent source data elements, s = 0 represents disconnection, and s = 1 represents connectivity;

[0024] Step 3.3, determine the distance d from the source data element to the target data element, and the amplitude compensation coefficients α and β;

[0025] In step 3.4, the source data sub-element label, the target data sub-element label, the distance transformation parameter d, the connectivity s, and the amplitude transformation parameters α and β are defined as a transformation compensation sequence.

[0026] Furthermore, the process flow analysis and prediction unit executes an analysis and prediction program, including data normalization processing, and performs process flow analysis on the normalized data. The normalization processing includes:

[0027] Step a: Select two sample data sets and define them as feature sample C and feature sample D respectively. Decompose feature sample C and feature sample D into Z subsamples. Each subsample is decomposed into two parts. The filter coefficient of the first part is defined as The second part of the filter coefficient is defined as C / D represents feature sample C or feature sample D, 1≤γ≤Z represents the γth decomposition;

[0028] Step b: For the first part of the filter coefficients of the γth decomposition, the first fusion criterion is used for fusion, and the feature sample C or the feature sample D is traversed, and the correlation of the feature sample C or the feature sample D is calculated to obtain the fusion weight. The first fusion criterion is:

[0029] Determine the P×Q window as region R, and calculate the gradient amplitude G of each feature sample sub-point in the horizontal and vertical directions within region R x [i,j] and G y [i,j], calculate the gradient value G(i,j)

[0030]

[0031] The inner product energy of the center point of region R is calculated as E(P(x,y))

[0032] <> is the intra-region product operation;

[0033] The correlation between feature sample C and feature sample D is calculated as:

[0034]

[0035] Assume the threshold is a and the weight coefficient is w C and w D for:

[0036] when When w C =0,w D =0;

[0037] hour, w D =1-w C ;

[0038] The calculated feature after the first part of the fusion is R F (x,y)=w C ·R C (x,y)+w C ·R C (x,y);

[0039] Step c: For the second part of the coefficients of the γth decomposition, the second fusion criterion is used to fuse them, and a window of P×Q is defined as the region R; the second fusion criterion is to calculate the maximum value E of the regional energy for fusion;

[0040] Where w(i,j) is the weight of each adjacent pixel in the area;

[0041] In step d, the fusion results of step b and step c are reconstructed using the inverse transformation corresponding to step a to obtain the normalized feature F.

[0042] Beneficial effects of the present invention: The present invention adopts hierarchical data transmission during the data collection and transmission required in the process of process analysis and configuration, thereby reducing transmission overhead and improving transmission efficiency. On this basis, the present invention selects to perform similarity judgment between the newly collected data and the historical data of the edge node. After similar data is judged, the similarity parameters of the two (i.e., the transformation compensation sequence) are directly used as the parameters to be transmitted for transmission. However, in order to prevent waste of overhead and reduce transmission efficiency, at this time, it is necessary to compare the size of the data segments before and after processing, and only use the new data segments for transmission after they are reduced. After receiving the data, the receiving end determines whether there are similarity parameters (i.e., the transformation compensation sequence) and solves the data through inverse operation. In this scheme, the data that has been transmitted to the receiving end is stored at the receiving end. If similar data exists, there is no need to retransmit the new data. Only the transformation relationship between the two is transmitted, which can greatly save transmission bandwidth and increase transmission efficiency. At the same time, the different types of data required for the process configuration analysis strategy are planned and processed to improve the efficiency of analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The present invention will be further described below with reference to the accompanying drawings and examples.

[0044] Figure 1 , Schematic diagram of the dynamic configuration method of equipment process flow.

[0045] Figure 2 ,Block diagram of rule-based process dynamic scheduling control method. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0047] Example 1

[0048] This embodiment provides a method for dynamically configuring an equipment process flow, which includes:

[0049] Step 1: The sensor collects the real-time status parameters of the process flow;

[0050] Step 2: The edge node receives the real-time status parameters of the process flow, which are defined as data to be processed;

[0051] Step 3: The edge node extracts the features of the data to be preprocessed, searches for similar data in the edge node's historical transmission database, records the similar data labels, and uses the similar data as reference data. If no similar data can be retrieved, the data to be preprocessed is defined as the data to be transmitted, and step 6 is directly executed.

[0052] Step 4: Using the reference data as the source and the data to be preprocessed as the target, calculate the transformation compensation sequence;

[0053] Step 5: Define the data to be transmitted by using similar data labels and transformation compensation sequences; compare the sizes of the data to be preprocessed and the data to be transmitted. If the data to be processed is smaller than the data to be transmitted, then execute step 6; otherwise, update the data to be preprocessed to the data to be transmitted.

[0054] Step 6: Determine the size of the data to be transmitted and the transmission threshold. If the data to be transmitted is smaller than the transmission threshold, return to step 2; otherwise, go to step 7.

[0055] Step 7: The edge node compresses and encodes the data to be transmitted and outputs it to the process flow configuration unit;

[0056] Step 8: The process flow configuration unit performs prediction analysis based on the received data and completes the process flow configuration.

[0057] In this embodiment, when performing the prediction analysis, the mathematical model of the manufacturing unit is first established:

[0058] The factory manufacturing unit is essentially a mixed assembly line. The minimum production cycle is usually used to describe a production cycle. The entire production is formed by a series of repeated operations of MPS. Establish a mathematical model of the manufacturing unit system: produce a series of products of varieties during the planning period. The demand for each variety is Dm. In a given M non-loop directed operation sequence graph G = (E, P), node E represents t i(i=1,2,3...n) operation element set, arc P represents the sequence relationship set between assembly operations, a node i∈E is annotated with a value representing the operation time, all varieties of operation sequence graphs are integrated into a comprehensive operation sequence graph, find a partition of the node set E Optimize specific objectives to meet certain job priorities, cycle times, and other constraints. The objective function is selected to minimize the number of workstations and the number of workstations. Given two main types of assembly line workstations and cycle times, the priority relationships of the job elements are merged into joint priority relationships, and the job time is converted into a comprehensive job time.

[0059] Second, the design of the manufacturing unit load balancing algorithm

[0060] The improved genetic hybrid algorithm is designed to solve the manufacturing unit load balancing problem, mainly including:

[0061] Initialization: Randomly generate the initial population and set relevant genetic parameters;

[0062] Fitness function evaluation: Establish a fitness function. The greater the fitness, the smaller the objective function, that is, the shorter the idle time and the more balanced the assembly line;

[0063] Selection operation: a method combining optimal preservation strategy and roulette wheel selection

[0064] Crossover operation: After selecting two individuals with the crossover probability Pc, perform single-point crossover according to the crossover method, then test whether the individuals after crossover meet the constraints, and then add the two best individuals to the new population;

[0065] Mutation operation: With the mutation probability Pm, any two gene segments are selected from the parent to exchange and obtain a new individual. After the mutation, the gene segment must also be tested to see if it meets the constraint conditions;

[0066] Algorithm termination condition: The algorithm ends after evolutionary generations.

[0067] Preferably, step 2 includes:

[0068] Step 2.1, define a feature extraction box with a width of w;

[0069] Step 2.2, use the feature extraction box to traverse the data to be preprocessed and complete feature extraction;

[0070] In step 2.3, the extracted features are matched with the historical data features in the edge node historical transmission database, and the historical data with a matching rate greater than a preset threshold is defined as similar data.

[0071] Preferably, step 3 comprises:

[0072] Step 3.1, determining the source data sub-element in the source data and the target data sub-element in the target data using the feature extraction frame as a unit;

[0073] Step 3.2, define the transformation equation as: E = d + η × s; the compensation equation is: I'' = αI' + β;

[0074] Wherein, η is the preset weight value, I' represents the source data element after deformation, and I'' represents the source data element after amplitude compensation; the connectivity s of adjacent source data elements, s = 0 represents disconnection, and s = 1 represents connectivity;

[0075] Step 3.3, determine the distance d from the source data element to the target data element, and the amplitude compensation coefficients α and β;

[0076] In step 3.4, the source data sub-element label, the target data sub-element label, the distance transformation parameter d, the connectivity s, and the amplitude transformation parameters α and β are defined as a transformation compensation sequence.

[0077] Preferably, the process flow analysis and prediction unit executes an analysis and prediction program, including data normalization processing, and performs process flow analysis on the normalized data, wherein the normalization processing includes:

[0078] Step a: Select two sample data sets and define them as feature sample C and feature sample D respectively. Decompose feature sample C and feature sample D into Z subsamples. Each subsample is decomposed into two parts. The filter coefficient of the first part is defined as The second part of the filter coefficient is defined as C / D represents feature sample C or feature sample D, 1≤γ≤Z represents the γth decomposition;

[0079] Step b: For the first part of the filter coefficients of the γth decomposition, the first fusion criterion is used for fusion, and the feature sample C or the feature sample D is traversed, and the correlation of the feature sample C or the feature sample D is calculated to obtain the fusion weight. The first fusion criterion is:

[0080] Determine the P×Q window as region R, and calculate the gradient amplitude G of each feature sample sub-point in the horizontal and vertical directions within region R x [i,j] and G y [i,j], calculate the gradient value G(i,j)

[0081]

[0082] The inner product energy of the center point of region R is calculated as E(P(x,y))

[0083] <> is the intra-region product operation;

[0084] The correlation between feature sample C and feature sample D is calculated as:

[0085]

[0086] Assume the threshold is a and the weight coefficient is w C and w D for:

[0087] when When w C =0,w D =0;

[0088] hour, w D =1-w C ;

[0089] The calculated feature after the first part of the fusion is R F (x,y)=w C ·R C (x,y)+w C ·R C (x,y);

[0090] Step c: For the second part of the coefficients of the γth decomposition, the second fusion criterion is used to fuse them, and a window of P×Q is defined as the region R; the second fusion criterion is to calculate the maximum value E of the regional energy for fusion;

[0091] Where w(i,j) is the weight of each adjacent pixel in the area;

[0092] In step d, the fusion results of step b and step c are reconstructed using the inverse transformation corresponding to step a to obtain the normalized feature F.

[0093] In addition, this embodiment also performs multi-objective optimization algorithm design.

[0094] The workshop production management module mainly includes SMT workshop production scheduling optimization, SMT workshop material management, workshop production plan management, and visualization of production management data and manufacturing data. Among them, SMT workshop production scheduling optimization technology breaks down the SMT workshop's production plan into specific production operations. In other words, it breaks down the company's existing processes, equipment, and planned orders into production operations to form an optimized workshop production scheduling plan.

[0095] Optimizing various production plans at the SMT workshop level is an important step in achieving scientific management and application of equipment, maximizing the production and processing capabilities of equipment, and improving its production efficiency and rapid response capabilities. The formulation of SMT workshop production plans is based on the company's existing processes, equipment, and planned orders. The following priority scheduling rules and optimization methods are designed:

[0096] SPT (Shortest Processing Time) rule: Prioritizes workpieces requiring the shortest processing time. This method minimizes average process time, reduces work-in-progress, and maximizes equipment utilization.

[0097] MWKR (Most Work Remaining) Rule: Prioritizes machining workpieces with the longest remaining processing time. This method minimizes the longest process time, reduces work-in-progress, and improves equipment utilization.

[0098] The EDD (Earliest Due Date) rule prioritizes processing workpieces with the earliest required completion date. This method minimizes the maximum delay (delay ≥ 0). This rule is often used for workpieces with strict delivery deadlines.

[0099] The SST (Shortest Slack Time) rule prioritizes processing workpieces with the shortest slack time. Slack time = completion deadline - current time - remaining processing time. This method is similar to the EDD rule, but better reflects the urgency of the task.

[0100] The Smallest Critical Ratio (SCR) rule prioritizes processing workpieces with the smallest critical ratio. Critical ratio = (Completion deadline - Current time) / Workpiece's remaining processing time. A smaller critical ratio indicates less slack time, reflecting the relative urgency of the task. This method and rule facilitates establishing a comparison standard for priorities among workpieces, resulting in superior overall performance.

[0101] At the same time, according to the current production resource conditions, a priority rule combination method is designed, such as SPT+MWKR+RANDOM.

[0102] Finally, for the production tasks of multiple varieties and variable batches, the execution of production plan generation, scheduling, change, etc. is monitored, tracked and recorded in real time, and a dynamic configuration algorithm for the process of network planning and rolling planning is designed.

[0103] When designing dynamic process configuration, for production tasks with multiple varieties and variable batches, the execution of production plan generation, scheduling, changes, etc. is monitored, tracked and recorded in real time, and dynamic process configuration algorithms for network plans and rolling plans are designed.

[0104] A network planning model consists of three basic elements: nodes, branches, and flows. Branches (arcs) represent the logical relationships (sequential order) between activities, nodes represent the individual activities (tasks) that comprise a project, and flows represent the parameters such as time, cost, and resources required to complete each activity. Once the network planning model is generated, steps such as process generation and dynamic adjustment can be performed.

[0105] Automatic process generation

[0106] After breaking down the current task into several independent activities and correctly drawing a network plan diagram according to the process logic relationship between each activity and the rules for drawing a network diagram, the various time parameters of the network plan can be calculated based on the duration of each activity. The duration of an activity is the time consumed to complete the activity with quality under specific conditions. The duration of each task depends on many factors such as labor organization, the technical or operational level of the staff. The calculation content of the engineering network plan time parameters is as follows: First, the minimum time required to complete the task (project duration); second, the possible start and end times of each activity in the project; and third, the maneuvering time allowed for each activity to be delayed without affecting the duration (activity time difference).

[0107] Process allocation and advancement

[0108] Based on the multi-level nested network plan of the network model, the product assembly task is broken down into various multi-layer network plans. The tasks are then refined layer by layer according to the recursive hierarchy. The network diagram of the previous layer can be recursively linked to the network diagram of the next layer, and the network diagram of the next layer can also be traced back to the network diagram of the previous layer. The entire network structure clearly reflects the process route and assembly level of the product assembly.

[0109] Dynamic configuration of processes

[0110] The objective environment of a shop floor is constantly changing. To adapt to this dynamic environment and improve planning accuracy, shop floor work plans must be adjusted in real time. This embodiment designs a rule-based dynamic process scheduling method that is applicable to a wide range of dynamically changing system environments as a global optimization method. During system operation, rules from a rule set are dynamically selected based on the actual state for scheduling control. This achieves better scheduling than that achieved by heuristic rules alone, at the expense of a reasonable amount of additional computation time.

[0111] exist Figure 2In this paper, FIFO and SPT methods are used for dynamic scheduling control of both input and output states. The input state variable is the average waiting time of a part in the queue. Dynamic selection: When the average waiting time of a part in the queue is less than a set value, the SPT rule is used for scheduling. If the waiting time of a part exceeds the set value, the FIFO rule is used for scheduling. Computational decision-making: When SPT is valid, the processing time of each part is calculated and the shortest one is selected for processing. When FIFO is valid, the arrival time of each part is calculated and the longest path is selected for processing.

[0112] This embodiment utilizes hierarchical data transmission during the data collection and transmission required for process analysis and configuration, reducing transmission overhead and improving transmission efficiency. Based on this, the present invention selects a similarity check between newly collected process data and historical data from edge nodes. Once similar data is determined, the similarity parameters (i.e., transformation compensation sequences) between the two are directly used as the parameters to be transmitted. However, to prevent wasted overhead and reduced transmission efficiency, the sizes of the pre- and post-processing data segments must be compared. Only when these are reduced will new data segments be used for transmission. Upon receiving the data, the receiving end determines whether similarity parameters (i.e., transformation compensation sequences) exist and decodes the data using inverse operations. In this solution, data already transmitted to the receiving end is stored at the receiving end. If similar data exists, new data is not retransmitted; only the transformation relationship between the two is transmitted. This significantly saves transmission bandwidth and increases transmission efficiency. Furthermore, the different types of data required for the process configuration analysis strategy are planned and processed to improve analysis efficiency.

[0113] Although the above describes the illustrative specific embodiments of the present invention so that those skilled in the art can understand the present invention, the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the appended claims, all inventions and creations based on the concepts of the present invention are protected.

Claims

1. A method for dynamic configuration of equipment process flow, characterized by: The dynamic configuration method of equipment process flow includes: Step 1: The sensor collects the real-time status parameters of the process flow; Step 2: The edge node receives the real-time status parameters of the process flow, which are defined as data to be pre-processed; Step 3: The edge node extracts the features of the data to be preprocessed, searches for similar data in the edge node's historical transmission database, records the similar data labels, and uses the similar data as reference data. If no similar data can be retrieved, the data to be preprocessed is defined as the data to be transmitted, and step 6 is directly executed. Step 4: Using the reference data as the source and the data to be preprocessed as the target, calculate the transformation compensation sequence; Step 5: Define the data to be transmitted by using similar data labels and transformation compensation sequences; compare the sizes of the data to be preprocessed and the data to be transmitted. If the data to be preprocessed is smaller than the data to be transmitted, then execute step 6; otherwise, update the data to be preprocessed to the data to be transmitted. Step 6: Determine the size of the data to be transmitted and the transmission threshold. If the data to be transmitted is smaller than the transmission threshold, return to step 2; otherwise, go to step 7. Step 7: The edge node compresses and encodes the data to be transmitted and outputs it to the process flow configuration unit; Step 8: The process flow configuration unit performs prediction analysis based on the received data and completes the process flow configuration; Step 2 includes: Step 2.1, define a feature extraction box with a width of w; Step 2.2, use the feature extraction box to traverse the data to be preprocessed and complete feature extraction; Step 2.3, matching the extracted features with the historical data features in the edge node historical transmission database, and defining the historical data with a matching rate greater than a preset threshold as similar data; Step 3 includes; Step 3.1, determining the source data sub-element in the source data and the target data sub-element in the target data using the feature extraction frame as a unit; Step 3.2, define the transformation equation as: E = d + η × s; the compensation equation is: I ‘’ =αI ′ +β; Wherein, η is the preset weight value, I ′ Represents the deformed source data element, I ‘’ represents the source data element after amplitude compensation; the connectivity s of adjacent source data elements, s = 0 means no connectivity, s = 1 means connectivity; Step 3.3, determine the distance d from the source data element to the target data element, and the amplitude compensation coefficients α and β; In step 3.4, the source data sub-element label, the target data sub-element label, the distance transformation parameter d, the connectivity s, and the amplitude transformation parameters α and β are defined as a transformation compensation sequence.

2. The method for dynamic configuration of equipment process flow according to claim 1, characterized in that: The process flow configuration unit executes an analysis and prediction program, including data normalization processing, and performs process flow analysis on the normalized data. The normalization processing includes: Step a: Select two sample data sets and define them as feature sample C and feature sample D respectively. Decompose feature sample C and feature sample D into Z subsamples. Each subsample is decomposed into two parts. The filter coefficient of the first part is defined as The second part of the filter coefficient is defined as C / D represents feature sample C or feature sample D, 1≤γ≤Z represents the γth decomposition; Step b: For the first part of the filter coefficients of the γth decomposition, the first fusion criterion is used for fusion, and the feature sample C or the feature sample D is traversed, and the correlation of the feature sample C or the feature sample D is calculated to obtain the fusion weight. The first fusion criterion is: Determine the P×Q window as region R, and calculate the gradient amplitude G of each feature sample sub-point in the horizontal and vertical directions within region R x [i,j] and G y [i,j], calculate the gradient value G(i,j) The inner product energy of the center point of region R is calculated as E(P(x,y)) <> is the intra-region product operation; The correlation between feature sample C and feature sample D is calculated as: Assume the threshold is a and the weight coefficient w C and w D for: when When C =0,w D =0; hour, w D =1-w C ; The calculated feature after the first part of the fusion is R F (x,y)=w C ·R C (x,y)+w C ·R C (x,y); Step c: For the second part of the coefficients of the γth decomposition, the second fusion criterion is used to fuse them, and a window of P×Q is defined as the region R; the second fusion criterion is to calculate the maximum value E of the regional energy for fusion; Where w(i,j) is the weight of each adjacent pixel in the area; In step d, the fusion results of step b and step c are reconstructed using the inverse transformation corresponding to step a to obtain the normalized feature F.

Images