| CPC G06F 11/0754 (2013.01) [B23Q 17/0957 (2013.01); G05B 19/4063 (2013.01); G05B 23/0221 (2013.01); G05B 23/0283 (2013.01); G06F 11/0736 (2013.01); G05B 2219/34477 (2013.01)] | 8 Claims |
|
1. A fault prediction system based on sensor data on a numerically controlled machine tool, comprising a controller, a flash memory and a plurality of sensors for collecting data on a running state of the numerical control machine tool as multi-channel data, a number of the plurality of sensors is p, and p is a positive integer greater than or equal to 2;
it is characterized in that, an output end of each of the sensors is connected to an input end of a multi-channel sensor interface circuit, and an output end of the multi-channel sensor interface circuit is connected to the controller;
the controller is also connected to a flash memory, and the flash memory comprises a construction module, a setting module, a superposition module, an exporting module, a limiting module, and a penalty item adding module, the controller is used to direct a processor to execute the construction module, the setting module, the superposition module, the exporting module, the limiting module, and the penalty item adding module program modules stored in the flash memory;
wherein the construction module is configured to form received p-channel data into tensor-data-one of p-channel, the tensor-data-one of p-channel comprises n number of sample data collected by the plurality of sensors, and each sample data comprises data collected by the plurality of the sensors at a certain sampling time, and data format of i-th sample data Yi(t) is: Yi(t)=[Yil(t), . . . , Yip(t)], t is sampling time of the i-th sample data; Yi(t) can be expressed by formula (1);
Yij(t)=Xij(t)+εij(t) (1)
Xij(t) is a signal function of collected data of jth channel, εij(t) is a noise function of the collected data of the jth channel, i is a positive integer less than or equal to N, j is a positive integer less than or equal to p, Yij(t) represents the collected data of the j-th channel at a sampling time t;
the setting module is configured to set data format of tensor-data-two Xi(t) of the p channel as: Xi(t)=[Xil(t), . . . , Xip(t)], and the data format of the tensor-data-two Xi(t) of the p channel is divided into L sample subspaces Sl, l=1, 2, . . . , L, L is a positive integer, in the same sample subspace, there is cross-correlation between the signal functions representing each signal, the signal functions of each signal in different sample subspaces have no correlation;
the superposition module is configured to set each sample subspace Sl as a linear superposition of Φl: Φl=[Φl1(t), . . . , Φldl(t)] of dl basis functions, as shown in formula (2):
 for all signal functions Xij(t), i=1, . . . , N, the sample subspace to which Xij(t) belongs remains unchanged, while the basis coefficients αij=[αij1, αij2 . . . , αijdl] corresponding to Xij(t) are different, and αijk is k-th component of the basis coefficient αij, where k is a positive integer less than or equal to dl;
wherein Xl={Xj(t)|Xj(t)∈Sl, j=1, . . . , p}, Xl is signal of respective sample subspaces, wherein q and dl are both positive integers, and αq is a set real number, R is a set of real numbers;
the exporting module is configured to obtain the formula (3) for each signal Xij(t)∈Xl:
 that is, Xij(t) is a signal function in Xl, which can be expressed as a linear combination of other pl−1 signals in this sample subspace, r is a positive integer, and bjr is a rth component of a sparse coefficient bj, wherein the sparse coefficient bj is configured to reflects the cross-correlation performance of the running state data of the numerical control machine tool, to implement equipment failure prediction of the numerically controlled machine tool;
the limiting module is configured to limit solution set by using the Lq norm to minimize an objective function for the solution of the sparse coefficient bj;
the penalty item adding module is configured to set that the collected data of the j-th channel has a total of S′−1 change points τs, s=1, . . . , S′−1, for each change point as a dividing point:
Yijs(t)=Xijs(t)+εij(t) (6)
among them, Xijs(b) is a signal function of the change point τs of the j-th channel, εij(t) is a noise function of the collected data of the j-th channel, Xijs (t) represents the change point τs of the j-th channel at the sampling time t, a penalty term is added to the sparsity coefficient bj, and S′ is a positive integer.
|