Classifications

• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
  • G06N7/00—Computing arrangements based on specific mathematical models
  • G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
  • G06F17/10—Complex mathematical operations
  • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing

Definitions

• the present invention relates to the field of the analysis of complex signals, in particular for the detection of certain particular patterns in signals of different natures.
• the analysis of these data is a real challenge to be able to value the data from these sensors.
• the analysis of a single signal does not pose a problem in itself, because this analysis is most often to treat the curve with known mathematical tools (eg calculation of derivatives, analysis of variability, etc.).
• the increase in heart rate alone may mean that the patient is running or is in cardiac distress;
• - Increasing the consumption of the industrial machine may mean that the carburation of the engine is not working well or that the operator has requested an increase in production rate.
• the present invention thus aims at a method of recognizing a pattern in a plurality of time signals of a different nature, the method comprising: / a / receiving the plurality of time signals;
• Ibl for each signal received from the plurality of signals, creating a temporal asynchronous signal including events, said events being the same forms of the asynchronous signal representative of a characteristic of the received signal;
• the activity profile comprises at least one activity value which decreases as a function of the elapsed time since a most recent event among the successive events of said asynchronous signal ; for a given time:
• Distances can be distances in the mathematical sense.
• the distances can be Euclidean distances, Manhattan distances, Minkoswski distances, Chebyshev distances or other distances.
• Ibl for each signal received from the plurality of signals, creating a temporal asynchronous signal including events, said events being the same forms of the asynchronous signal representative of a characteristic of the received signal;
• Ib'l use the events of each asynchronous signal as current events and use the standard contexts of the first level of the hierarchical model as typical common contexts;
• the activity profile comprises at least one activity value which decreases as a function of the elapsed time since a most recent event among the successive events of said asynchronous signal ; Idl for each given time among a plurality of given times:
• M determination of a context, said context being defined as the set of values of the activity profiles, at the given time, of the asynchronous signals created,
• the first hierarchical level makes it possible to identify very basic forms, the higher order hierarchical levels making it possible to identify more complex forms from the lower levels.
• the comparison of the step / f / may comprise the calculation of a Bhattacharyya distance. Indeed, it has been found experimentally that this distance allowed better quality recognition.
• the comparison of the step / f / may include the calculation of a standardized distance.
• the comparison of step / f / may be a comparison between a number of instances of determination of the typical contexts at the last occurrence of step I02I with a signature base types.
• the decay of the activity profile can be a function of the level of the current hierarchical model.
• This adaptation may allow a slower rate of decay for higher hierarchical levels in order to have a responsiveness of these lower levels. Conversely, low level hierarchical levels (ie the first ones used) may have higher responsiveness.
• a computer program, implementing all or part of the method described above, installed on a pre-existing equipment, is in itself advantageous, since it allows pattern recognition in a plurality of signals.
• FIGS. 5a and 5b described in detail below can form the flowchart of the general algorithm of such a computer program.
• FIG. 1b illustrates a generation of events from a signal representative of a respiratory rate
• FIGS. 2a and 2b are examples of "activity profile" for the events resulting from received signals
• FIGS. 3a and 3b are examples of representation of a context determined from three received signals
• FIGS. 4a to 4d are examples of representation of four typical contexts
• FIG. 4e is an example of generation of four event flows from recognition in a signal of the typical contexts of FIGS. 4a to 4d;
• FIG. 4f is an example of determination of a signature for four event streams
• FIG. 5a is an example of a flow chart of a method according to a possible embodiment of the invention (i.e. without using a hierarchical model of typical contexts);
• FIG. 5b is an example of a flow chart of a method according to another possible embodiment of the invention (i.e. using a hierarchical model of standard contexts);
• FIG. 6 illustrates a device for implementing an embodiment according to the invention.
• Figure 1a illustrates generation of events from a signal representative of a heart rhythm.
• the curve 101 is representative of a heart rhythm.
• This curve can be received as continuous data (eg analog signal) or as sampled data (eg digital signal).
• the variability of a signal pattern eg variability of the R wave repetition period in an electrocardiogram or variability of an engine cycle time for a machine tool
• these characteristics can be representative of a function of the signal (eg of the derivative of the signal or of any mathematical transformation of the signal).
• an "asynchronous" signal comprising markers or events (eg curve 102 or 103). These events are most often Diracs in the asynchronous signal, because their generation is simple. Nevertheless, these events can be any pattern (eg triangular signal, rectangular, sinusoidal portion, AC signal portion, etc.). Each event of said asynchronous signal is advantageously the same pattern or a similar pattern (their amplitude or polarity (ie their meaning relative to the zero value) can nevertheless change from one event to another).
• the curve 102 is an asynchronous signal whose events are generated during the appearance of a QRS complex in the electrocardiogram of the curve 101.
• the curve 103 is an asynchronous signal whose events are generated when the width of a QRS complex of the electrocardiogram of the curve 101 exceeds a target value (eg average over the previous x minutes).
• the curve 105 is an asynchronous signal whose events are generated when the respiratory rate exceeds a predetermined threshold (i.e., horizontal dashed line) in the curve 104.
• the curve 106 is an asynchronous signal whose events are generated when the derivative of the respiratory rate exceeds a target value in the curve 104.
• FIGS. 2a and 2b are examples of three "activity profiles" t ⁇ S, for the events resulting from three received signals pi, p 2 and p 3 .
• This exponential decay can be illustrated in Figure 2b (see event 220, for example).
• the nature and the parameters of this decay can be chosen and different for each analyzed signal.
• the value of the activity profile S can be set to the sum (possibly weighted) of the current value of S just before the event 222 (ie h 0 ) and h.
• the decay of the curve S will start from the value h + h 0 as shown in FIG . 2b .
• T (p, i) max (t / ) ⁇ j ⁇ i or
• p ⁇ T (p, t) defines a map of the times of the last events occurring temporally just before a reference time (i.e. t).
• a predetermined time constant ⁇ S may be any function decreasing with time t over an interval comprising as lower bound T (p, t)).
• the function p ⁇ S (p, i) is called the "context" of the input signals for a given time t.
• the order of the components of the vector can be arbitrary in a large number of situations, it is possible to establish an order among these components as a function of the spatial and / or temporal dynamics of the signals.
• two consecutive components in a given context can represent signals having a near spatial and / or temporal dynamics.
• Figures 4a to 4d are examples of representation of four typical contexts. In order to detect patterns in the signals, it is possible to define standard contexts that will be compared with the previously defined S contexts.
• Figure 3a can be compared with each of the typical contexts 401, 402, 403 and 404.
• FIG. 4e it is also possible (FIG. 4e) to generate events for new asynchronous signals when a typical context is retained / recognized:
• the hierarchical model may have only one level in a particular case, it is possible to define a signature.
• Figure 4f is an example of determining a signature for four event streams.
• the signature is defined as being the vector or the histogram 420 of the number of occurrence (respectively n 40 i, n 40 2, n 40 3, n 404 ) of each of the determined standard contexts (respectively 401, 402 , 403, 404) for the last occurrence of the method described above.
• the distance between two histograms K x and ⁇ 2 (P ⁇ r example 420 and 421) can be calculated as a mathematical distance between two vectors whose coordinates are the number of occurrences for each of the standard contexts: It is also possible to calculate a standardized distance as follows:
• the distance from Bhattacharyya can also be used to replace the classic distance: with W j ( ⁇ ) the number of occurrences of the i th standard context of the histogram 3 ⁇ 4 ⁇ .
• the signature may also be the general form of the asynchronous signals of the four event streams.
• Each of the standard signatures may be associated with a given pattern in the received signals: for example, the type signature 421 may correspond to a mechanical failure of an engine while the type signature 422 corresponds to a normal operation of said engine.
• Fig. 5a is an exemplary flowchart of a method according to a possible embodiment of the invention (i.e. without using hierarchical model of typical contexts).
• step 502 On receiving signals 501, it is possible to create (step 502) asynchronous signals by generating events on the basis of signal characteristic detections, as described with reference to FIGS. 1a and 1b. Once these asynchronous signals have been created, it is possible to create (step 503), from each created asynchronous signal, an activity profile as described with reference to FIGS. 2a and 2b.
• the determined context can be defined as a vector of the values of the activity profiles, at the given time, of the asynchronous signals created.
• a typical context from a set of predetermined standard contexts (for example, pre-computed and stored in a database 505) by minimizing the distance between the context determined in step 504 and the typical contexts.
• a "pattern" 509 in the plurality of received signals Based on the determined typical context, it is then possible to determine (step 508) a "pattern" 509 in the plurality of received signals.
• Fig. 5b is an exemplary flowchart of a method according to another possible embodiment of the invention (i.e. using hierarchical model of typical contexts).
• step 502 On receiving signals 501, it is possible to create (step 502) asynchronous signals by generating events on the basis of signal characteristic detections, as described with reference to FIGS. 1a and 1b.
• the determined context can be defined as a vector of the values of the activity profiles, at the given time, of the asynchronous signals created. This determination can be performed for a plurality of given times (eg this plurality of times being a sampling of the time with a predetermined step).
• Each recognition of a standard context may allow the generation (step 512) of an event as described in connection with FIG. 4e.
• asynchronous signal ie in addition to new asynchronous signals created
• an asynchronous signal from a step similar to step 502.
• a received signal is a fast signal (whose variations are notable for a predetermined period of time)
• the signal received is slow (its variations are negligible over the predetermined period of time)
• This integration of this asynchronous signal will be done in parallel with the asynchronous signals generated in step 512 for the "faster" signals already integrated.
• step 506 determines a "pattern" by comparing the typical contexts determined at the last occurrence of step 506 with a signature base types.
• This comparison may comprise, for example, the calculation of the number of occurrences of determination of the standard contexts at the last occurrence of step 506 as described with reference to FIG. 4f.
• the comparison may also be a distance comparison between the asynchronous signals generated at the last occurrence of step 513 (in this case step 513 must be executed when the last level of the hierarchical model is used) and a signature database.
• Fig. 6 shows an exemplary pattern recognition device in one embodiment of the invention.
• the device comprises a computer 600, comprising a memory 605 for storing instructions for implementing the method, the received measurement data, and temporary data for performing the various steps of the method as described above. .
• the computer further comprises a circuit 604.
• This circuit may be, for example: a processor capable of interpreting instructions in the form of a computer program.
• a programmable electronic chip such as an FPGA chip (for "Field- Programmable Gâte Array "in English) whose steps of the method of the invention are described, for example, in a VHDL or Verilog code (hardware description languages for representing the behavior and the architecture of an electronic system).
• This computer has an input interface 603 for receiving the signals to be analyzed, and an output interface 606 for providing the recognized pattern.
• the computer may include, for easy interaction with a user, a screen 601 and a keyboard 602.
• the keyboard is optional, particularly in the context of a computer in the form of a touch pad, for example.
• FIG. 5a or 5b is a typical example of a program, some of whose instructions can be carried out with the device described. As such, FIG. 5a or 5b may correspond to the flowchart of the general algorithm of a computer program within the meaning of the invention.
• the proposed method can be used to analyze any type of data / signal in any possible industrial domain, without any particular limitation.

Landscapes

• Engineering & Computer Science (AREA)
• Physics & Mathematics (AREA)
• General Physics & Mathematics (AREA)
• Data Mining & Analysis (AREA)
• Theoretical Computer Science (AREA)
• Mathematical Analysis (AREA)
• Computational Mathematics (AREA)
• Mathematical Physics (AREA)
• Mathematical Optimization (AREA)
• Pure & Applied Mathematics (AREA)
• Algebra (AREA)
• Probability & Statistics with Applications (AREA)
• Software Systems (AREA)
• General Engineering & Computer Science (AREA)
• Operations Research (AREA)
• Evolutionary Biology (AREA)
• Databases & Information Systems (AREA)
• Life Sciences & Earth Sciences (AREA)
• Bioinformatics & Cheminformatics (AREA)
• Bioinformatics & Computational Biology (AREA)
• Artificial Intelligence (AREA)
• Computing Systems (AREA)
• Evolutionary Computation (AREA)
• Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
• Management, Administration, Business Operations System, And Electronic Commerce (AREA)
• Measuring And Recording Apparatus For Diagnosis (AREA)
• Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
• Image Analysis (AREA)
• Complex Calculations (AREA)

Applications Claiming Priority (2)

Publications (1)

Family

ID=57680373

Family Applications (1)

Country Status (7)

Family Cites Families (16)

• 2017
  • 2017-09-08 WO PCT/FR2017/052393 patent/WO2018046868A1/fr unknown
  • 2017-09-08 CA CA3035520A patent/CA3035520A1/fr active Pending
  • 2017-09-08 JP JP2019513319A patent/JP7209622B2/ja active Active
  • 2017-09-08 AU AU2017322448A patent/AU2017322448B2/en active Active
  • 2017-09-08 US US16/331,808 patent/US11080523B2/en active Active
  • 2017-09-08 CN CN201780055286.5A patent/CN109844739B/zh active Active
  • 2017-09-08 EP EP17771501.8A patent/EP3510495A1/de active Pending
• 2022
  • 2022-10-20 JP JP2022168637A patent/JP2023017794A/ja not_active Ceased

Also Published As

Similar Documents

Legal Events