CN110215581A - Breathing Suppotion device hardware fault detection method - Google Patents
Breathing Suppotion device hardware fault detection method Download PDFInfo
- Publication number
- CN110215581A CN110215581A CN201910254742.XA CN201910254742A CN110215581A CN 110215581 A CN110215581 A CN 110215581A CN 201910254742 A CN201910254742 A CN 201910254742A CN 110215581 A CN110215581 A CN 110215581A
- Authority
- CN
- China
- Prior art keywords
- breathing suppotion
- value
- detection method
- fault detection
- adf
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61M—DEVICES FOR INTRODUCING MEDIA INTO, OR ONTO, THE BODY; DEVICES FOR TRANSDUCING BODY MEDIA OR FOR TAKING MEDIA FROM THE BODY; DEVICES FOR PRODUCING OR ENDING SLEEP OR STUPOR
- A61M16/00—Devices for influencing the respiratory system of patients by gas treatment, e.g. mouth-to-mouth respiration; Tracheal tubes
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M99/00—Subject matter not provided for in other groups of this subclass
- G01M99/005—Testing of complete machines, e.g. washing-machines or mobile phones
Abstract
The present invention discloses a kind of Breathing Suppotion device hardware fault detection method, chooses more Breathing Suppotion equipment, and wherein 70%-99% is normal device, remaining is faulty equipment；Acquire the data of the Breathing Suppotion equipment, including actual pressure value P, actual flow value F, flow sensor AD value ADP, pressure sensor AD value ADF and secondary speed R；It is recommended that linear mapping relation；Establish the BP neural network of a deep learning.Compared with the relevant technologies, Breathing Suppotion device hardware fault detection method of the invention, detects timeliness and accuracy is higher.
Description
Technical field
The present invention relates to the field of medical instrument technology more particularly to a kind of Breathing Suppotion device hardware fault detection methods.
Background technique
If the Breathing Suppotion equipment of noninvasive ventilator one kind is a kind of barotropic gas that persistently exports to the equipment of user, user
Typically last for when in use a few hours even a couple of days, if the critical component turbine of Breathing Suppotion equipment, pressure sensor,
Flow sensor breaks down, and it is poor to will lead to the curative effect that user is treated using equipment, in some instances it may even be possible to cause life danger
Danger.
Based on this, current Breathing Suppotion equipment is usually that the self-test of a hardware is carried out when being switched on and being powered, from
Inspection allows user to use again after passing through.
The Breathing Suppotion equipment of the relevant technologies only carries out hardware fault self-test when being powered, in process of self-test, even if having
Fine difference also can be by detection, and part hidden failure or complex fault can not be found when self-test, in operation one
It can just break down after the section time, thus be easy to cause and adverse effect is caused to user.
To which the Breathing Suppotion device hardware fault detection method of the relevant technologies there is detection timeliness and accuracy is lower
Deficiency.
Therefore, it is necessary to provide a kind of new Breathing Suppotion device hardware fault detection method solution above-mentioned technical problem.
Summary of the invention
The purpose of the present invention is overcoming above-mentioned technical problem, a kind of detection timeliness and the higher breathing branch of accuracy are provided
Holding equipment hardware fault detection method.
The present invention provides a kind of Breathing Suppotion device hardware fault detection method, comprising:
More Breathing Suppotion equipment are chosen, wherein 70%-99% is normal device, remaining is faulty equipment；
Acquire the data of the Breathing Suppotion equipment, including actual pressure value P, actual flow value F, flow sensor AD value
ADP, pressure sensor AD value ADF and secondary speed R；
It is recommended that linear mapping relation, is mapped as 1 for the result of normal device, faulty equipment is mapped as 0, hasLinear Mapping formula are as follows:
F (P, F, ADP, ADF, R)=w1*P+w2*F+w3*ADP+w4*ADF+w5*R+b.Wherein, w1, w2, w3, w4, w5
For the corresponding weight coefficient of each parameter, b is correction factor；
The BP neural network of a deep learning is established, input layer is (P, F, ADP, ADF, R), and middle layer has 1 layer, 5
Node, output layer is any one numerical value between 0~1, at each node, using Sigmoid as excitation function, table
Up to formula are as follows:Wherein, z=wX+b, w=(w1, w2, w3, w4, w5), X=(P, F, ADP, ADF, R), excitation
Output of the result of function as layer where the neuron, passes to next layer of neuron, then in output layer y=F0
(W0FH(WH*X+bH)+b0Final output value is calculated, wherein F0For the 0th layer of excitation function, FHFor the excitation function of middle layer,
W0For initial random weight, b0For initial random correction factor, WHAnd bHThe respectively calculating weight and correction factor of middle layer, F0
=f (P, F, ADP, ADF, R) feeds back neural network using gradient descent method after obtaining y, then carries out to w coefficient
Amendment, the global error until exporting result reach minimum, stop neural network at this time, export final w as neural network
Training result.
Preferably, the method also includes:
Acquire the data in vector X；
According toI=0,1,2 ..., N calculate the average value of corresponding data, and N is the total number of data；
W vector sum b vector is brought into (P, F, ADP, ADF, R) and is calculated, final output result is obtained.
Preferably, acquiring the data in vector X is the data acquired in the 300ms of air-breathing end in vector X.
Preferably, when exporting result more than or equal to 0.5, judge that hardware effort is normal, otherwise judgement carries out hardware inspection
It looks into.
Preferably, when exporting result less than 0.5, Breathing Suppotion device prompts user is broken down, by corresponding data
And maintenance record storage re-starts training after increasing a plurality of maintenance record to training result database, generates new weight
Value.
Preferably, 20 maintenance records of every increase, re-start training, generate new weighted value.
Preferably, the Breathing Suppotion number of devices of selection is greater than 100.
Preferably, in the Breathing Suppotion equipment of selection, 95% is normal device, and 5% is faulty equipment.
Compared with prior art, Breathing Suppotion device hardware fault detection method provided by the invention, by choosing more
Breathing Suppotion equipment and the data for acquiring the more Breathing Suppotion equipment, establish linear mapping relation, resettle deep learning
Neural network makes Breathing Suppotion equipment by way of machine learning, just can be carried out inspection when hardware occurs the omen of failure
It surveys, it is more timely also more accurate to the detection of hardware fault.
Specific embodiment
The technical scheme in the embodiments of the invention will be clearly and completely described below, it is clear that described implementation
Example is only a part of the embodiments of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, this field is common
Technical staff's all other embodiment obtained without making creative work belongs to the model that the present invention protects
It encloses.
The hardware composition of Breathing Suppotion equipment includes following key hardware: turbine, flow sensor, pressure sensing
Device.Wherein turbine is the source of gas-pressurized, and normal work when should export pressure value set by user；Flow sensing
Device and pressure sensor read the flow and pressure data of Breathing Suppotion equipment gas outlet respectively, for analysis user's the present situation with
And turbine control provides foundation, the numerical value that normal work when reads should be consistent with actual numerical value.
When turbine soon breaks down, common performance is the same pressure and flow of output, and turbine is corresponding
Revolving speed is higher；When flow sensor and pressure sensor soon break down, it is common performance be identical flow or
Pressure, corresponding AD value are relatively low or higher.
To which the present invention provides a kind of Breathing Suppotion device hardware fault detection method, comprising:
More Breathing Suppotion equipment are chosen, wherein 70%-99% is normal device, remaining is faulty equipment；
Acquire the data of the Breathing Suppotion equipment, including actual pressure value P, actual flow value F, flow sensor AD value
ADP, pressure sensor AD value ADF and secondary speed R；
It is recommended that linear mapping relation, is mapped as 1 for the result of normal device, faulty equipment is mapped as 0, hasLinear Mapping formula are as follows:
F (P, F, ADP, ADF, R)=w1*P+w2*F+w3*ADP+w4*ADF+w5*R+b.Wherein, w1, w2, w3, w4, w5
For the corresponding weight coefficient of each parameter, b is correction factor；
The BP neural network of a deep learning is established, input layer is (P, F, ADP, ADF, R), and middle layer has 1 layer, 5
Node, output layer is any one numerical value between 0~1, at each node, using Sigmoid as excitation function, table
Up to formula are as follows:
Wherein, z=wX+b, w=(w1, w2, w3, w4, w5), X=(P, F, ADP, ADF, R), excitation
Output of the result of function as layer where the neuron, passes to next layer of neuron, then in output layer y=F0
(W0FH(WH*X+bH)+b0Final output value is calculated, wherein F0For the 0th layer of excitation function, FHFor the excitation function of middle layer,
W0For initial random weight, b0For initial random correction factor, WHAnd bHThe respectively calculating weight and correction factor of middle layer, F0
=f (P, F, ADP, ADF, R) feeds back neural network using gradient descent method after obtaining y, then carries out to w coefficient
Amendment, the global error until exporting result reach minimum, stop neural network at this time, export final w as neural network
Training result.
In present embodiment, the method also includes:
Acquire the data in vector X；
According toI=0,1,2 ..., N calculate the average value of corresponding data, and N is the total number of data；
W vector sum b vector is brought into (P, F, ADP, ADF, R) and is calculated, final output result is obtained.
Specifically, the data in acquisition vector X are the data acquired in the 300ms of air-breathing end in vector X.
Specifically, judging that hardware effort is normal when exporting result more than or equal to 0.5, otherwise judgement carries out hardware inspection
It looks into.
More specifically, when exporting result less than 0.5, Breathing Suppotion device prompts user is broken down, will be corresponding
Data and maintenance record, which are stored to training result database, re-starts training after increasing a plurality of maintenance record, generates new power
Weight values.
More specifically, 20 maintenance records of every increase, re-start training, generate new weighted value.
In present embodiment, the Breathing Suppotion number of devices of selection is greater than 100, preferably 2000.
Specifically, 95% is normal device, and 5% is faulty equipment in the Breathing Suppotion equipment chosen.
The following are specific embodiments:
Firstly, choosing 2000 equipment, there is 95% normal device in such devices, there is 5% faulty equipment.
Secondly, acquiring the data of all 2000 equipment, these data include actual pressure value P, actual flow value F, stream
Quantity sensor AD value ADP, pressure sensor AD value ADF, secondary speed R.
Again, it is proposed that the result of normal device is mapped as 1 by linear mapping relation, and faulty equipment is mapped as 0, that is, hasSpecifically, Linear Mapping formula are as follows: f (P, F, ADP, ADF, R)=
w1*P+w2*F+w3*ADP+w4*ADF+w5*R+b.Wherein, w1, w2, w3, w4, w5 are the corresponding weight coefficients of each parameter, and b is
Correction factor.
Finally, establishing the BP neural network of a deep learning, input layer is (P, F, ADP, ADF, R), and middle layer shares 1
Layer, has 5 nodes, output layer is any one numerical value between 0~1.At each node, using Sigmoid as excitation
Function, expression formula are as follows:Z=wX+b, w=(w1, w2, w3, w4, w5), X=(P, F, ADP, ADF, R).
The result of excitation function can pass to next layer of neuron as the output of layer where the neuron.Then in output layer y
=F0(W0FH(WH*X+bH)+b0Final output value is calculated, wherein F0Indicate the 0th layer of excitation function, FHIndicate swashing for middle layer
Encourage function, W0Indicate initial random weight, b0Indicate initial random correction factor, WHAnd bHIndicate middle layer calculating weight and
Correction factor, in the present invention F0=f (P, F, ADP, ADF, R).After obtaining y, using gradient descent method to neural network
It is fed back, then w coefficient is modified, the global error until exporting result reaches minimum, stop neural network at this time,
Export training result of the final w as neural network.
2000 groups of data are fully entered and carry out machine learning in neural network, w1, w2, w3, w4 can be obtained, w5's
Data, then in use, as long as Breathing Suppotion equipment continues to monitor the evaluation energy of f (P, F, ADP, ADF, R)
Whether tell will break down.
For Breathing Suppotion equipment when monitoring, data when needing to choose more steady as far as possible are for failure point
The numerical operation of analysis a little or continuously takes a little because if taking at random, it is possible to meet random error, and then there may be accidentally
Sentence.In order to avoid these problems, Breathing Suppotion equipment can choose the data of statistical collection end-tidal, and user is typically at this time
Quiescent stage, the movement to breathe no more substantially, flow and pressure are also substantially at steady state, are most suitable for acquiring data.
Based on the above analysis, Breathing Suppotion equipment is in the end exhaled every time, that is, before air-breathing starts next time
In a bit of time, the data in vector X are acquired.Specifically, being all data before acquiring air-breathing within 300ms, then count
The average value of corresponding data is calculated, i.e.,I=0,1,2 ..., N, wherein N indicates the total number of data.Obtaining average value
Afterwards, just w vector sum b vector is also brought into f (P, F, ADP, ADF, R) and is calculated, obtain final output result.
When exporting result more than or equal to 0.8, illustrates that device hardware work is normal, is not in problem,
When being greater than or equal to 0.5 and less than 0.8, Breathing Suppotion equipment can prompt user to pay attention to maintaining, when less than 0.5
It waits, it is that equipment the omen of hardware fault occurs that Breathing Suppotion equipment, which will be considered that, is checked or replaced accessory.
Breathing Suppotion equipment remind user break down when, can by corresponding data back to tranining database,
Then after maintenance personal is confirmed whether failure, the data and corresponding result are stored in tranining database, no matter setting
It is whether accurate for what is determined, then after 20 maintenance records of every increase, training is re-started, generates new weighted value, thus
Training data can constantly be increased, then get more accurate weight, promote the judgement accuracy of Breathing Suppotion equipment.
Compared with prior art, Breathing Suppotion device hardware fault detection method provided by the invention, by choosing more
Breathing Suppotion equipment and the data for acquiring the more Breathing Suppotion equipment, establish linear mapping relation, resettle deep learning
Neural network makes Breathing Suppotion equipment by way of machine learning, just can be carried out inspection when hardware occurs the omen of failure
It surveys, it is more timely also more accurate to the detection of hardware fault.
The above description is only an embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hair
Equivalent structure or equivalent flow shift made by bright specification is applied directly or indirectly in other relevant technical fields,
Similarly it is included within the scope of the present invention.
Claims (8)
1. a kind of Breathing Suppotion device hardware fault detection method characterized by comprising
More Breathing Suppotion equipment are chosen, wherein 70%-99% is normal device, remaining is faulty equipment；
Acquire the data of the Breathing Suppotion equipment, including actual pressure value P, actual flow value F, flow sensor AD value ADP,
Pressure sensor AD value ADF and secondary speed R；
It is recommended that linear mapping relation, is mapped as 1 for the result of normal device, faulty equipment is mapped as 0, hasLinear Mapping formula are as follows:
F (P, F, ADP, ADF, R)=w1*P+w2*F+w3*ADP+w4*ADF+w5*R+b.Wherein, w1, w2, w3, w4, w5 are each
The corresponding weight coefficient of parameter, b are correction factor；
Establish the BP neural network of a deep learning, input layer is (P, F, ADP, ADF, R), and middle layer has 1 layer, 5 nodes,
Output layer is any one numerical value between 0~1, at each node, using Sigmoid as excitation function, expression formula
Are as follows:
Wherein, z=wX+b, w=(w1, w2, w3, w4, w5), X=(P, F, ADP, ADF, R), excitation function
Output of the result as layer where the neuron, next layer of neuron is passed to, then in output layer y=F0(W0FH
(WH*X+bH)+b0Final output value is calculated, wherein F0For the 0th layer of excitation function, FHFor the excitation function of middle layer, W0For
Initial random weight, b0For initial random correction factor, WHAnd bHThe respectively calculating weight and correction factor of middle layer, F0=f
(P, F, ADP, ADF, R) feeds back neural network using gradient descent method, then repairs to w coefficient after obtaining y
Just, reach minimum until exporting the global error of result, stop neural network at this time, export final w as neural network
Training result.
2. Breathing Suppotion device hardware fault detection method according to claim 1, which is characterized in that the method is also wrapped
It includes:
Acquire the data in vector X；
According toThe average value of corresponding data is calculated, N is the total number of data；
W vector sum b vector is brought into (P, F, ADP, ADF, R) and is calculated, final output result is obtained.
3. Breathing Suppotion device hardware fault detection method according to claim 2, which is characterized in that in acquisition vector X
Data be the data acquired in the 300ms of air-breathing end in vector X.
4. Breathing Suppotion device hardware fault detection method according to claim 3, which is characterized in that when output result is big
When 0.5, judge that hardware effort is normal, otherwise judgement carries out hardware check.
5. Breathing Suppotion device hardware fault detection method according to claim 4, which is characterized in that when output result is small
When 0.5, Breathing Suppotion device prompts user is broken down, and corresponding data and maintenance record are stored to training result data
Library re-starts training after increasing a plurality of maintenance record, generates new weighted value.
6. Breathing Suppotion device hardware fault detection method according to claim 5, which is characterized in that every to increase by 20 dimensions
Record is repaired, training is re-started, generates new weighted value.
7. Breathing Suppotion device hardware fault detection method according to claim 1, which is characterized in that the breathing branch of selection
Holding equipment number is greater than 100.
8. Breathing Suppotion device hardware fault detection method according to claim 7, which is characterized in that the breathing branch of selection
In holding equipment, 95% is normal device, and 5% is faulty equipment.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910254742.XA CN110215581A (en) | 2019-03-31 | 2019-03-31 | Breathing Suppotion device hardware fault detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910254742.XA CN110215581A (en) | 2019-03-31 | 2019-03-31 | Breathing Suppotion device hardware fault detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110215581A true CN110215581A (en) | 2019-09-10 |
Family
ID=67822491
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910254742.XA Pending CN110215581A (en) | 2019-03-31 | 2019-03-31 | Breathing Suppotion device hardware fault detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110215581A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111110979A (en) * | 2019-12-31 | 2020-05-08 | 湖南明康中锦医疗科技发展有限公司 | Self-checking system and method for water tank of respiratory support equipment |
| WO2021092845A1 (en) * | 2019-11-14 | 2021-05-20 | Elekta (Shanghai) Technology Co., Ltd. | Predictive maintenance of dynamic leaf guide based on deep learning |
-
2019
- 2019-03-31 CN CN201910254742.XA patent/CN110215581A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021092845A1 (en) * | 2019-11-14 | 2021-05-20 | Elekta (Shanghai) Technology Co., Ltd. | Predictive maintenance of dynamic leaf guide based on deep learning |
| CN111110979A (en) * | 2019-12-31 | 2020-05-08 | 湖南明康中锦医疗科技发展有限公司 | Self-checking system and method for water tank of respiratory support equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110215581A (en) | Breathing Suppotion device hardware fault detection method | |
| CN109740977A (en) | The evaluation method of Gas Generator Set carbon emission influence factor based on grey correlation analysis | |
| CN109597396A (en) | A kind of distribution transforming on-line fault diagnosis method based on high amount of traffic and transfer learning | |
| CN109670714A (en) | Ship gas turbine synthetical condition assessment method based on degree of membership analysis | |
| CN109856494A (en) | A kind of Diagnosis Method of Transformer Faults based on support vector machines | |
| CN108388237A (en) | A kind of method for diagnosing faults, device, equipment and the medium of discrete manufacturing equipment | |
| CN105956755A (en) | Method and system for establishing quantitative relationship of general power line loss rate influencing factors | |
| CN102114292B (en) | Detection method, device and system of medical device | |
| CN116562623A (en) | Real-time planting risk assessment system suitable for velvet mushroom is planted | |
| CN106446570A (en) | Conference influence assessment method based on academic big data | |
| CN106598026B (en) | Power transmission and transformation equipment state overhauling time detection method and device | |
| CN109559019A (en) | A kind of Electrical Power System Dynamic safety evaluation method based on risk index | |
| Osigwe et al. | Integrated gas turbine system diagnostics: components and sensor faults quantification using artificial neural networks | |
| CN105279553B (en) | A kind of height adds to water system fault degree recognition methods | |
| CN110956330A (en) | Method and system for predicting line loss of power transmission line based on multidimensional influence quantity | |
| CN109886428A (en) | A kind of power equipment safety configuration check method | |
| CN109581194A (en) | A kind of electronic system malfunction Test Strategy dynamic creation method | |
| CN105759782B (en) | A kind of car fault diagnosis construction of strategy method based on density of infection | |
| Billinton et al. | Wind power modelling and the determination of capacity credit in an electric power system | |
| CN107301508A (en) | A kind of gyroscope damage quantitative appraisal procedure based on expert's comprehensive assessment | |
| Kim et al. | A quantitative approach to modeling the information flow of diagnosis tasks in nuclear power plants | |
| CN115619287B (en) | Multi-source data fusion-based hydroelectric generating set state degradation evaluation method and system | |
| Chu et al. | A relaxed support vector data description algorithm based fault detection in distribution systems | |
| CN115358531B (en) | Virtual power plant operation risk analysis method and device | |
| CN109726461A (en) | A kind of multiplexing electric abnormality judgment method based on user power utilization data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190910 |
|
| WD01 | Invention patent application deemed withdrawn after publication |