US 2021/0364482 A1 · PREDICTION METHOD OF PART SURFACE ROUGHNESS AND TOOL WEAR BASED ON MULTI-TASK LEARNING

Abstract

A prediction method of part surface roughness and tool wear based on multi-task learning belong to the file of machining technology. Firstly, the vibration signals in the machining process are collected; next, the part surface roughness and tool wear are measured, and the measured results are corresponding to the vibration signals respectively; secondly, the samples are expanded, the features are extracted and normalized; then, a multi-task prediction model based on deep belief networks (DBN) is constructed, and the part surface roughness and tool wear are taken as the output of the model, and the features are extracted as the input to establish the multi-task DBN prediction model; finally, the vibration signals are input into the multi-task prediction model to predict the surface roughness and tool wear.

A prediction method of part surface roughness and tool wear condition based on multi-task learning, comprising the following steps:

step one:
collecting vibration signal during machining
installing tri-axial acceleration sensor at the position of spindle close to tool holder to collect the tri-vibration signal of the spindle during the machining process of the CNC machine tool, and the vibration signal of the machine tool load is intercepted from the tri-axial vibration signal
step two:
measuring the part surface roughness and tool wear
using measurement equipment to measure the part surface roughness and tool wear;
when measuring the part surface roughness, dividing the surface of the parts into equal intervals according to the sampling length, and measuring a roughness value in each interval, and whether the surface quality of the parts is qualified or not is divided according to the roughness;
tool wear condition measurement, tool wear is measured every fixed cutting distance, and the tool is divided into three conditions:
normal, worn and damaged according to the tool whether there is obvious wear and damage;
finally, dividing the vibration data collected in the machining process by the length of vibration data corresponding to each sampling length, and the surface roughness label and tool wear condition label are respectively corresponding to the vibration data after division;
step three:
sample expansion and feature extraction
(1) sample expansion
according to equation (1), adding M degree of Gaussian white noise to the collected vibration signal;
V
i
s
=
(
k
2
⁢
i
+
k
)
⁢
σ
M
⁢
V
st
s
(
1
)
where:
Vsts is the original vibration signal, S is the sample index, k is the different intensity parameter, i=1, 2, . . . , M is used to control the different noise level, σ is the sample standard variance;
after adding noise, each sample is expanded to M+1 group sample {vsts, v1s, . . . , VMs}, the sample expansion is as follows:
Sa=Vsts+Vis (2)
(2) dynamic signal feature extraction and normalization
in this method, normalizing the variance, root mean square, kurtosis, impulse factor and skewness features extracted from equation (3)˜(7);
Va
=
1
N
-
1
⁢
∑
t
=
1
N
⁢
(
V
i
s
⁡
(
t
)
-
V
¯
)
2
(
3
)
RMS
=
(
1
N
⁢
∑
t
=
1
N
⁢
V
i
s
⁡
(
t
)
2
)
1
/
2
(
4
)
KUR
=
[
1
N
⁢
∑
t
=
1
N
⁢
(
V
i
s
⁡
(
t
)
-
V
¯
)
4
]
/
[
1
N
⁢
∑
t
=
1
N
⁢
(
V
i
s
⁡
(
t
)
-
V
¯
)
2
]
2
(
5
)
IF
=
[
V
i
s
⁡
(
t
)
max
-
V
i
s
⁡
(
t
)
min
]
/
[
1
N
⁢
∑
t
=
1
N
⁢

V
i
s
⁡
(
t
)

]
(
6
)
Ske
=
[
1
N
⁢
∑
t
=
1
N
⁢
(
V
i
s
⁡
(
t
)
-
V
¯
)
3
]
/
[
1
N
⁢
∑
t
=
1
N
⁢
(
V
i
s
⁡
(
t
)
-
V
¯
)
2
]
3
/
2
(
7
)
where:
Vis(t) is the vibration data of a sample, t is the index of vibration data, N is the number of data contained in each sample, and v is the average value of each sample;
step four:
the construction and training of multi-task prediction model based on improved DBN
(1) construction of improved deep belief networks
constructing the deep neural network A based on the deep belief networks;
the deep neural network A is composed of three layers of restricted Boltzmann machine and a BP network;
among them, the restricted Boltzmann machine is an energy-based model, and its joint probability distribution is specified by the energy function, and the definition of the energy function is shown in equation (8),
E
⁡
(
v
,
h
⁢
;
⁢
⁢
θ
)
=
-
∑
α
=
1
V
⁢
∑
β
=
1
H
⁢
w
α
⁢
β
⁢
v
α
⁢
h
β
-
(
∑
α
=
1
V
⁢
b
α
⁢
v
α
+
∑
β
=
1
H
⁢
a
β
⁢
h
β
)
(
8
)
among them, vα and hβ are the binary states of visible and hidden units, θ={w, b, a} are the parameters of the model;
wαβ are the weight between visible units α and hidden units β, bα and aβ are offsets of visible units α and hidden units β, respectively, V and H is the number of visible and hidden units;
the energy of hidden neurons is as follows:
E
β
=
∑
α
=
1
V
⁢
w
α
⁢
β
⁢
v
α
+
b
α
(
9
)
similarly, the energy of visible neurons is as follows:
E
α
=
∑
β
=
1
H
⁢
w
α
⁢
β
⁢
h
β
+
a
β
(
10
)
marking the input layer to the first layer restricted Boltzmann machine as B1, and from the second layer restricted Boltzmann machine to the output layer as B2, copying B2 and marking it as B3, B3 is connected with the first layer restricted Boltzmann machine of deep neural network A, at this time, arranging the network structure B2 and B3 side by side and connecting with B1;
according to the task requirements, setting the number of output layer nodes of B2 and B3 to complete the improvement of deep belief network;
the improved deep belief network B is composed of network structure B1, B2 and B3, which can realize the function of multi-task learning;
(2) training of improved deep belief network
firstly, using the greedy algorithm to pre-training the restricted Boltzmann machine of each layer in the deep neural network A, and assigning the weight of the trained deep neural network A to the improved deep belief networks B;
then, based on the back-propagation algorithm, the supervised fine-tuning of the improved deep belief network B is carried out, and the loss function as equation (11) is adjusted to the optimal value to obtain multi-task predicting model;
L
⁡
(
θ
)
=
-
∑
j
=
1
W
⁢
λ
j
m
[
∑
k
=
1
m
⁢
y
^
k
⁢
⁢
log
⁢
⁢
y
k
]
(
11
)
among them, W is the total number of tasks in the multi-task model, λj is the weight of each task, m is the total number of fine-tuning samples, ŷk is the actual label, yk is the prediction label;
step five:
predicting the part surface roughness and tool wear condition
in the actual machining process, pre-processing the vibration data collected and inputting the pre-processed vibration data into the multi-task prediction model after training;
the model outputs the part surface roughness and tool wear condition;
task 1:
outputs 1 when the part surface roughness is qualified, otherwise it is 0;
task 2:
output 0 when the tool is not worn, output 1 when it is worn, and output 2 when it is damaged.

425 more elements

Prediction method of part surface roughness and tool wear based on multi-task learning

Life of this patent

Prediction method of part surface roughness and tool wear based on multi-task learning

Life of this patent

Abstract

Claims1