Summary of the invention
The objective of the invention is to consider the problems referred to above and a kind of high-precision test result that can obtain is provided, and then obtain the online test method of surface machining accuracy of complex parts of the machining accuracy of processed curved surface.The present invention has overcome complex-curved needs and has moved to the weak point that CMM upward could obtain the high Precision Detection result; Can be directly on CNC milling machine, the part with complex-curved characteristic be carried out the online detection of shape surface accuracy; Obtain the high-precision test result through error compensation; Finally obtain the shape face machining accuracy of part, improved production efficiency effectively.
Technical scheme of the present invention is: the online test method of the surface machining accuracy of complex parts of the present invention includes following steps:
1) contact is triggered on the main shaft that gauge head is installed in CNC milling machine;
2) tested curved surface is detected path planning;
3) drive CNC milling machine, obtain the coordinate of the survey ball center (3) at gauge head and curved surface contact point to be measured place one by one; Surveying ball is the parts that contact triggers gauge head, and its shape is a ball with the high accuracy of manufacture and high rigidity, is installed on the gauge head main body and with tested curved surface directly to contact;
4) coordinate of above-mentioned survey ball center is surveyed radius of a ball compensation, gauge head pretravel error compensation and machine tool error compensation, thereby obtain the testing result of measuring point high precision; Wherein machine tool error need adopt the machine tool error detecting instrument to obtain;
5) the desirable cad model with testing result and part compares analysis, finds each measuring point deviation corresponding with cad model, thereby obtains the machining accuracy of curved surface to be measured.
Above-mentioned survey radius of a ball compensation, gauge head pretravel error compensation need calculate the direction of normal of curved surface measuring point; For gauge head pretravel error compensation; Also need calculate the pretravel error of gauge head at each measuring point direction of normal, compensation method is to adopt to predict based on the neutral net of RBF (RBF).
The used machine tool error detecting instrument of above-mentioned machine tool error compensation is a laser interferometer.
The deviation that above-mentioned measuring point is corresponding with cad model adopts the computational methods of the minimum range of point to curved surface to find the solution.
It is following that above-mentioned neutral net based on RBF RBF is carried out forecast method:
Regularization RBF network is that network is approached in a kind of three layers of feedforward part with single hidden layer, and verified, it and BP network can both approach any continuous function with arbitrary accuracy; And, compare the BP network, its training time is shorter; And it satisfies the approximate error of sample and the flatness of approximating curve simultaneously; In practice, the supervised training of network can be regarded a kind of process of curve match as, utilizes regularization RBF algorithm; Through training, realize the Nonlinear Mapping between the input and output space to network;
Regularization RBF topology of networks is made up of the output layer of a radially basic neuronic latent layer, a linear neuron, and the input point quantity of network is N, and latent number of nodes is P, and output node quantity is l; The latent node number of network equals to import sample number, and all are imported the center that samples are made as RBF, and each RBF is got unified expansion constant;
RBF achieved by the input ?
to the output
mapping,?
(
) using the radial basis function as either a hidden node activation function, the choice of Gauss function as the radial basis function; W of output power matrix, where
(
;
) for the hidden layer
to the output node Level
nodes between synaptic weights; linear activation function as the output layer neurons;
According to regularization RBF Principles of Network, the training process of RBF is: RBF neutral net input and output variable are confirmed in (1), promptly with the detection side to input node as network, corresponding pretravel error is the output node of network; (2) form training set network is trained, be i.e. some groups of teacher's data of picked at random from detected pretravel error information as network; (3) input forecast sample, with the network that trains predict any detection side to the pretravel error, with remaining measuring point data as forecast sample.
The method that the computational methods of the minimum range of above-mentioned point to curved surface are found the solution is following: curved surface is divided into enough little grid, the calculating measuring point (
x m,
y m,
z m) to the distance of total-grid node, all these minimum value and value are exactly the minimum range of point to curved surface.
The present invention is because after being employed in complex-curved machining to be detected; The method that on the workbench of CNC milling machine, directly detects; Obtain the high-precision test result thereby the data of obtaining are carried out error compensation, this testing result is analyzed and then obtained the machining accuracy of tested curved surface.Advantage of the present invention is: method of the present invention can directly carried out online detection to complex-curved after machining on the CNC milling machine; Through the data of obtaining being surveyed radius of a ball compensation, pretravel error compensation and machine tool error compensation; Obtain the high-precision test result; And then obtain the machining accuracy of processed curved surface; Overcome complex-curved needs and moved to the weak point that CMM upward could obtain the high Precision Detection result, improved production efficiency effectively, the present invention has remarkable economic efficiency, social benefit.The present invention is that a kind of design is ingenious, function admirable, the online test method of the surface machining accuracy of convenient and practical complex parts.
The specific embodiment
Embodiment:
The online test method of the surface machining accuracy of complex parts of the present invention,
The online test method of the surface machining accuracy of complex parts of the present invention includes following steps:
1) contact is triggered on the main shaft that gauge head 4 is installed in CNC milling machine;
2) tested curved surface 2 is detected path planning;
3) drive CNC milling machine, obtain the coordinate of the survey ball center 3 at gauge head and curved surface contact point to be measured place one by one; Surveying ball is the parts that contact triggers gauge head, and its shape is a ball with the high accuracy of manufacture and high rigidity, is installed on the gauge head main body and with tested curved surface directly to contact;
4) coordinate of above-mentioned survey ball center 3 is surveyed radius of a ball compensation, gauge head pretravel error compensation and machine tool error compensation, thereby obtain the testing result of measuring point high precision; Wherein machine tool error need adopt the machine tool error detecting instrument to obtain;
5) the desirable cad model with testing result and part compares analysis, finds each measuring point deviation corresponding with cad model, thereby obtains the machining accuracy of curved surface to be measured.
Above-mentioned survey radius of a ball compensation, gauge head pretravel error compensation need calculate the direction of normal of curved surface measuring point; For gauge head pretravel error compensation; Also need calculate the pretravel error of gauge head at each measuring point direction of normal, compensation method is to adopt to predict based on the neutral net of RBF (RBF).
The used machine tool error detecting instrument of above-mentioned machine tool error compensation is a laser interferometer.
The deviation that above-mentioned measuring point is corresponding with cad model adopts the computational methods of the minimum range of point to curved surface to find the solution.
It is following that above-mentioned neutral net based on RBF (RBF) is carried out forecast method:
Regularization RBF network is that network is approached in a kind of three layers of feedforward part with single hidden layer, and verified, it and BP network can both approach any continuous function with arbitrary accuracy; And, compare the BP network, its training time is shorter; And it satisfies the approximate error of sample and the flatness of approximating curve simultaneously; In practice, the supervised training of network can be regarded a kind of process of curve match as, utilizes regularization RBF algorithm; Through training, realize the Nonlinear Mapping between the input and output space to network;
Regularization RBF topology of networks is made up of the output layer of a radially basic neuronic latent layer, a linear neuron, and the input point quantity of network is N, and latent number of nodes is P, and output node quantity is l; The latent node number of network equals to import sample number, and all are imported the center that samples are made as RBF, and each RBF is got unified expansion constant;
RBF achieved by the input
to the output
mapping,?
(
) using the radial basis function as either a hidden node activation function, the choice of Gauss function as the radial basis function; W of output power matrix, where
(
;
) for the hidden layer
a node to the output layer
nodes between the synaptic weights; linear activation function as the output layer neurons;
According to regularization RBF Principles of Network, the training process of RBF is: RBF neutral net input and output variable are confirmed in (1), promptly with the detection side to input node as network, corresponding pretravel error is the output node of network; (2) form training set network is trained, be i.e. some groups of teacher's data of picked at random from detected pretravel error information as network; (3) input forecast sample, with the network that trains predict any detection side to the pretravel error, with remaining measuring point data as forecast sample.
The method that the computational methods of the minimum range of above-mentioned point to curved surface are found the solution is following: curved surface is divided into enough little grid, the calculating measuring point (
x m,
y m,
z m) to the distance of total-grid node, all these minimum value and value are exactly the minimum range of point to curved surface.
The embodiment of the invention is to utilize CNC milling machine that complex-curved workpiece is carried out the high-precision test method; This method is adapted to when CNC milling machine workpiece 11 processed; The workpiece 11 of the embodiment of the invention is a complex curved surface parts; After workpiece 11 is accomplished a manufacturing procedure, directly on the workbench of CNC milling machine, workpiece 11 is detected, realize processing and detecting and all on CNC milling machine, carry out; Can avoid that workpiece 11 is moved to other checkout equipments (like three coordinate measuring machine) and go up the second positioning error that detection brings, also can avoid size and the big workpiece 11 of weight are carried the inconvenience that is brought.Workpiece 11 to be detected in the method for present embodiment directly detects on the workbench of CNC milling machine after machining, and may further comprise the steps:
Step 1: contact is triggered on the main shaft that gauge head 4 is installed in CNC milling machine.The contact that drives main shaft triggers gauge head 4 motions, and contact triggers gauge head 4 and implements the coordinate of tested curved surface 2 is detected, and testing result is recorded in and detects in the software.
Step 2: tested curved surface 2 is detected path planning.Utilize detection software that tested curved surface is carried out points planning and detects path planning.
Step 3: drive CNC milling machine, obtain the coordinate that contact triggers the survey ball center 3 at gauge head 4 and tested curved surface contact point 5 places one by one;
Step 4: above-mentioned survey ball center 3 coordinates are surveyed radius of a ball compensation, contact triggering gauge head 4 pretravel error compensations and machine tool error compensation, thereby obtain the testing result of measuring point high precision;
As shown in Figure 1, the key of surveying radius of a ball compensation is to obtain the direction of normal 1 of measured point, utilizes formula to survey radius of a ball compensation then
(1)
In the following formula, (
X,
Y,
Z) be the coordinate of contact point A, (
x,
y,
z) be the coordinate of surveying the B of ball center,
rBe the survey radius of a ball,
nBe measuring point per unit system vector.
The compensation aspect of gauge head pretravel error 7; As shown in Figure 2; From contact trigger gauge head 4 contact workpiece surfaces to triggering signal produce during this period of time in, contact triggers gauge head 4 additional movements slight distance, the general title because the caused error amount of this section slight distance is a gauge head pretravel error.Need utilize standard ball to detect the pretravel error of gauge head on all directions, and utilize RBF (RBF) algorithm predicts to go out the pretravel error 7 on any direction of normal 1.
Aspect the machine tool error compensation; According to kinematic principle; Object has six-freedom degree along a certain rectilinear motion; Promptly three translation freedoms and three revolution frees degree just have six geometric error components, promptly along the straightness error of three mutual vertical direction with to the rotation error of three mutual vertical direction.The three-axis numerical control milling machine generally all has (X, Y, Z) three orthogonal linear motion axis.Therefore for the three-axis numerical control milling machine, three axles in edge move and co-exist in 18 error components, add between three axles also to have the error of perpendicularity, always have 21 error components.Present embodiment utilizes laser interferometer to measure this 21 error components.
Measuring point method arrow, the pretravel sum of errors machine tool error of above-mentioned acquisition are input in the detection software systems, and the survey ball center coordinate that the numerical control system feedback is returned is compensated.
Present embodiment uses TP6L gauge head system.
Step 5: the desirable cad model of testing result and part is compared analysis, find each measuring point deviation corresponding, thereby obtain the machining accuracy of curved surface to be measured with cad model.
Need carry out points planning (being the arrangement mode of measuring point on curved surface) in the above-mentioned steps two, and calculate the direction of normal 1 of each measuring point, detect path planning then.Said method is vowed to vowing on the curved surface of measuring point place and through the method for said tested point.
Contact triggers gauge head 4 and at the uniform velocity contacts the measuring point on the tested curved surface along direction of normal 1 in the above-mentioned steps three; In this process; Contact triggering gauge head 4 is at a high speed located moving direction 8 to 6 from gauge head along the detection side and is moved to gauge head low speed direction of closing 9; The preparatory contact distance 10 of gauge head utilizes software systems to receive the centre coordinate information that contact triggers gauge head 4 in contact process among the figure.
In the above-mentioned steps four, need utilize standard ball to detect the pretravel error of gauge head on all directions, and utilize RBF (RBF) algorithm predicts to go out the pretravel error on any direction of normal.Utilize laser interferometer measurement to obtain the geometric error of lathe.