Methods and systems for performing root cause analysis with a trained machine learning model
Methods and systems of using a trained machine-learning model to perform root cause analysis on a manufacturing process. A pre-trained machine learning model is provided that is trained to predict measurements of non-faulty parts. The pre-trained model is trained on training measurement data regarding physical characteristics of manufactured parts as measured by a plurality of sensors at a plurality of manufacturing stations. With the trained model, then measurement data from the sensors is received regarding the manufactured part and the stations. This new set of measurement data is back propagated through the pre-trained model to determine a magnitude of absolute gradients of the new measurement data. The root cause is then identified based on this magnitude of absolute gradients. In other embodiments the root cause is identified based on losses determined between a set of predicted measurement data of a part using the model, and actual measurement data.
1 . A computer-implemented method of utilizing a pre-trained machine learning model to perform root cause analysis on a manufacturing process, the method comprising:
providing a pre-trained machine learning model that is trained to predict measurements of non-faulty parts, wherein the pre-trained machine learning model is trained based on a first set of measurement data regarding physical characteristics of a first plurality of manufactured parts as measured by a plurality of sensors at a plurality of manufacturing stations;
receiving, from the plurality of sensors at the plurality of manufacturing stations, a second set of measurement data regarding physical characteristics of a second plurality of manufactured parts and an identification of the plurality of manufacturing stations;
back propagating the second set of measurement data through the pre-trained machine learning model to determine a magnitude of absolute gradients of the second set of measurement data; and
identifying a root cause within the manufacturing process based on the magnitude of absolute gradients.
2 . The method of claim 1 , further comprising:
training a binary classification model using outputs of the pre-trained machine learning model to output predictions of whether at least one part type of a plurality of part type will be faulty;
determining a magnitude of absolute gradients of the binary classification model; and
identifying at least one of the plurality of part types as a root cause within the manufacturing setting based on the magnitude of absolute gradients.
3 . The method of claim 1 , wherein the root cause is identified as
R xk =arg max(|∂ f θ ([ x 1 , . . . x k−1 ,S 1 , . . . ,S k−1 ],T x )/∂ x 1 ,x 2 , . . . ,x k−1 ,S 1 , . . . ,S k−1 |)
wherein f θ is the pre-trained machine learning model, x is a portion of the second set of measurement data regarding physical characteristics of the second plurality of manufactured parts, S is the identification of the plurality of manufactured stations, and k is one of the plurality of manufacturing stations in which a fault is identified.
4 . The method of claim 3 , wherein the pre-trained model is trained by:
via a time-series dynamics machine learning model, encoding the first set of measurement data into a latent space having a plurality of nodes, each node associated with the first set of measurement data of one of the non-faulty parts as measured at one of the manufacturing stations.
5 . The method of claim 4 , wherein the pre-trained model is further trained by:
via a prediction machine learning model, determining a predicted measurement of a first of the non-faulty parts at a first of the plurality of manufacturing stations based on the latent space of at least some of the first set of measurement data not including the measurement data corresponding to the first manufactured part at the first manufacturing station.
6 . The method of claim 5 , wherein the pre-trained model is further trained by:
via the prediction machine learning model, comparing the predicted measurement of the first manufactured part to the measurement data of the first manufactured part at the first manufacturing station; and
based on a difference between the predicted measurements and the actual measurement data, updating parameters of the machine learning model until convergence.
7 . The method of claim 1 , wherein the plurality of sensors includes image sensors or laser measurement sensors.
8 . A computer-implemented method of utilizing a pre-trained machine learning model to perform root cause analysis on a manufacturing process, the method comprising:
providing a pre-trained machine learning model that is trained to predict measurements of non-faulty parts, wherein the pre-trained machine learning model is trained based on a first set of measurement data regarding physical characteristics of a first plurality of manufactured parts as measured by a plurality of sensors at a plurality of manufacturing stations;
utilizing the pre-trained machine learning model to produce a set of predicted measurement data for each manufacturing station;
receiving, from the plurality of sensors at the plurality of manufacturing stations, a second set of measurement data regarding physical characteristics of a second plurality of manufactured parts and an identification of the plurality of manufacturing stations;
determining losses between the set of predicted measurement data and the second set of measurement data for each of the plurality of manufacturing stations; and
identifying a root cause within the manufacturing process based on the losses.
9 . The method of claim 8 , wherein the root cause includes a portion of the second measurement data regarding physical characteristics of the second plurality of manufactured parts, and wherein the root cause is represented by R xk =x i* i* j ;
wherein
i
*
,
i
j
*
=
arg
max
0
≤
i
<
k
,
j
i
(
L
(
f
θ
(
[
{
x
1
1
,
x
1
2
,
…
,
x
1
j
1
}
,
…
,
{
x
i
1
,
…
,
x
i
j
i
}
]
,
(
s
1
,
…
,
s
i
)
,
T
x
)
,
x
i
+
1
)
)
;
wherein f θ is the pre-trained machine learning model, x is a portion of the second set of measurement data regarding physical characteristics of the second plurality of manufactured parts, S is the identification of the plurality of manufacturing stations, T is a type of the second plurality of manufactured parts, and k is one of the plurality of manufacturing stations in which a fault is identified.
10 . The method of claim 8 , further comprising:
wherein the pre-trained model is trained by:
via a time-series dynamics machine learning model, encoding the first set of measurement data into a latent space having a plurality of nodes, each node associated with the first set of measurement data of one of the non-faulty parts as measured at one of the manufacturing stations.
11 . The method of claim 10 , wherein the pre-trained model is further trained by:
via a prediction machine learning model, determining a predicted measurement of a first of the non-faulty parts at a first of the plurality of manufacturing stations based on the latent space of at least some of the first set of measurement data not including the measurement data corresponding to the first manufactured part at the first manufacturing station.
12 . The method of claim 11 , wherein the wherein the pre-trained model is further trained by:
via the prediction machine learning model, comparing the predicted measurement of the first manufactured part to the measurement data of the first manufactured part at the first manufacturing station; and
based on a difference between the predicted measurements and the actual measurement data, updating parameters of the machine learning model until convergence.
13 . The method of claim 8 , wherein the plurality of sensors includes image sensors or laser measurement sensors.
14 . A computer-implemented method of utilizing a pre-trained machine learning model to perform root cause analysis to determine a part type causing a fault within a manufacturing setting, the method comprising:
providing a pre-trained machine learning model that is trained to predict measurements of non-faulty parts, wherein the pre-trained machine learning model is trained based on a first set of measurement data regarding physical characteristics of a first plurality of manufactured parts as measured by a plurality of sensors at a plurality of manufacturing stations;
training a binary classification model using outputs of the pre-trained machine learning model to output predictions of whether at least one part type of a plurality of part type will be faulty;
determining a magnitude of absolute gradients of the binary classification model; and
identifying at least one of the plurality of part types as a root cause within the manufacturing setting based on the magnitude of absolute gradients.
15 . The method of claim 14 , further comprising:
receiving, from the plurality of sensors at the plurality of manufacturing stations, a second set of measurement data regarding physical characteristics of a second plurality of manufactured parts and an identification of the plurality of manufacturing stations;
back propagating the second set of measurement data through the pre-trained machine learning model to determine a magnitude of absolute gradients of the second set of measurement data; and
identifying a root cause within the manufacturing process based on the magnitude of absolute gradients.
16 . The method of claim 14 , wherein the root cause is represented by R P =T P* , wherein T is a type of part P, and where
P
*
=
arg
max
T
∈
T
p
(
i
)
,
1
≤
i
≤
n
∑
i
=
1
n
∂
g
φ∪θ
(
[
x
1
(
i
)
,
…
,
x
k
(
i
)
]
,
(
s
1
(
i
)
,
…
s
k
(
i
)
)
,
T
x
)
∂
T
wherein x is a portion of the measurement data regarding physical characteristics of the plurality of manufactured parts, S is the identification of the plurality of manufactured stations, and k is one of the plurality of manufacturing stations in which a fault is identified.
17 . The method of claim 14 , wherein the pre-trained model is trained by:
via a time-series dynamics machine learning model, encoding the first set of measurement data into a latent space having a plurality of nodes, each node associated with the first set of measurement data of one of the non-faulty parts as measured at one of the manufacturing stations.
18 . The method of claim 17 , wherein the pre-trained model is further trained by:
via a prediction machine learning model, determining a predicted measurement of a first of the non-faulty parts at a first of the plurality of manufacturing stations based on the latent space of at least some of the first set of measurement data not including the measurement data corresponding to the first manufactured part at the first manufacturing station.
19 . The method of claim 18 , wherein the wherein the pre-trained model is further trained by:
via the prediction machine learning model, comparing the predicted measurement of the first manufactured part to the measurement data of the first manufactured part at the first manufacturing station; and
based on a difference between the predicted measurements and the actual measurement data, updating parameters of the machine learning model until convergence.
20 . The method of claim 14 , wherein the plurality of sensors includes image sensors or laser measurement sensors.