Service overview
Machine learning, predictive maintenance, computer vision, sensor analytics and forecasting. This service is part of that division and is scoped individually around what you need to learn, decide or publish.
Sensors turn physical quantities into signals, and signals carry information mixed with noise, artefacts and the effects of the measurement chain itself. Signal processing separates what is wanted from what is not, and extracts features that describe the underlying behaviour: a vibration spectrum that reveals a bearing defect, an acoustic emission burst that marks crack growth, a time-frequency pattern that distinguishes operating modes.
The techniques are mature, and the risk lies in applying them without understanding their assumptions. A filter that removes noise can also remove signal. A spectrum computed from a short or non-stationary record can mislead. A feature that separates two states in one dataset may fail when sensor placement changes. Careful signal analysis starts from the physics of the source and the characteristics of the sensor and acquisition system.
If you plan to publish the outcome, see manuscript & publication support, for example computational & simulation manuscripts or original research article writing.
What this service covers
Problems this service can address
- Noise and interference masking the phenomena of interest
- Choosing sampling rates and filters without distorting signals
- Features that do not generalise between machines or tests
Methods and analytical approaches
- Filtering and denoising design
- Fourier, wavelet and time-frequency analysis
- Feature extraction and selection
- Sensor-fusion concepts
Applications
- Vibration and acoustic signal analysis
- Biomedical and physical signal studies
- Test-rig data processing
Relevant engineering disciplines
Scope, inputs and deliverables
Scope of work
- Data audit and definition of the prediction or decision task
- Model development with structure-aware validation and baselines
- Evaluation, error analysis and uncertainty reporting
- Documentation sufficient for others to reproduce the work
- Feasibility check on: raw signals with sampling information
Typical inputs required from you
- Raw signals with sampling information
- Sensor type and placement description
- Events or states to detect
Deliverables
- Processing pipeline description
- Extracted features and spectral plots
- Interpretation and recommendations
Highly specialised requests are first checked for feasibility from this information. Please do not send confidential data through the public form; describe it first and agree secure transfer afterwards.
Project workflow and method
- 01Enquiry and requirement assessment
- 02Technical scoping and proposal
- 03Execution
- 04Quality review and delivery, with an agreed round of revisions
Read the full project workflow.
In depth: how the work is done
Acquisition, sampling and sensor characteristics
Quality begins at acquisition. Sampling rates must satisfy the Nyquist criterion with margin and be accompanied by anti-aliasing filters. Sensor bandwidth, sensitivity, mounting resonance and noise floor determine what can be measured. Time synchronisation across channels matters for multi-sensor analyses. We review the acquisition chain, since problems at this stage cannot be fixed afterwards, and record parameters so that the processing can be interpreted correctly.
Filtering and denoising
Filters are designed for a purpose: low-pass to remove high-frequency noise, high-pass to remove drift, band-pass to isolate a frequency range, notch to remove mains interference. Filter type and order trade sharpness against phase distortion and transient behaviour; zero-phase filtering avoids delay in offline analysis. Denoising approaches such as wavelet thresholding or adaptive filters suit non-stationary or broadband noise. Effects of the filter on the feature of interest are checked.
Spectral and time-frequency analysis
Fourier analysis reveals the frequency content of stationary signals, with windowing and averaging choices affecting leakage and variance. For non-stationary signals such as transients and speed-varying machinery, short-time Fourier, wavelet and related time-frequency methods show how content evolves. Order analysis uses rotation-synchronous sampling to track components tied to shaft speed. Envelope analysis demodulates high-frequency resonances excited by repetitive impacts, a standard approach for bearing diagnostics.
Feature extraction and selection
Features summarise signals in quantities for diagnosis or modelling: statistical measures, spectral band energies, harmonic amplitudes, entropy and others. Features should relate to the physics of the fault or phenomenon and be robust to changes in operating conditions. Feature selection reduces redundancy and overfitting, and uses validation that respects the data structure. Each feature's behaviour across operating conditions is examined, not just its separation of classes in one dataset.
Sensor fusion and validation
Combining several sensors can improve reliability and observability, through simple rules, statistical fusion or state estimation. Fusion requires aligned data and understanding of each sensor's uncertainty and failure modes. Validation compares extracted features and conclusions against independent information, such as inspection results or controlled fault tests. Conclusions are phrased in terms of what the signals support, with uncertainty.
Quality control and good practice
Quality-control approach
- Data are audited for quality, leakage and drift before modelling.
- Validation respects data structure (by specimen, machine or time) and includes simple baselines.
- Performance is reported with uncertainty and failure cases, not only headline scores.
- Model limits and the conditions under which it should not be trusted are documented.
Common pitfalls this service helps avoid
- Aliasing from inadequate sampling. Frequencies above half the sampling rate fold into the measured band and cannot be removed afterwards.
- Over-filtering. Aggressive filters can remove or distort the signal features of interest.
- Spectral estimates from short records. Insufficient averaging or non-stationarity makes spectra unreliable.
- Features sensitive to operating conditions. Load and speed changes can alter features and be mistaken for faults.
What good reporting looks like
A signal-analysis report describes the sensors and acquisition parameters, the data and its quality, the processing steps with parameters and justification, the spectral or time-frequency results with figures, the extracted features and their physical meaning, the validation against independent information, uncertainty and limitations. Scripts and parameters are documented so results can be reproduced, and the report states which conclusions depend on assumptions about sensors or operating conditions.
An illustrative scenario
Hypothetical example written for explanation. It is not a client case, a result or a testimonial.
Imagine a researcher analysing vibration from a gearbox test rig. The raw spectrum is dominated by gear mesh harmonics, masking a weak fault signature. After resampling in the angle domain to remove speed fluctuation effects and applying an envelope analysis around a structural resonance, a series of sidebands appears at a frequency matching the fault geometry. The researcher confirms the finding with a controlled test in which a known defect is introduced and reports how the feature changes with load.
Questions to settle before you begin
- Is the sampling rate and bandwidth adequate for the phenomenon?
- What noise and interference sources are present?
- Do the features change with speed or load?
- What independent information can validate my interpretation?
- Are multiple sensors synchronised?
Key terms
- Nyquist frequency
- Half the sampling rate, the highest frequency that can be represented without aliasing.
- Aliasing
- Misrepresentation of high frequencies as lower ones due to insufficient sampling.
- Envelope analysis
- Demodulation of a signal to reveal repetitive impacts.
- Wavelet transform
- A time-frequency method using scaled and shifted wavelets.
- Order analysis
- Analysis of signals relative to rotation speed rather than time.
Working together on AI, Machine Learning & Data Science projects
AI and data science engagements start with the decision the model is meant to support and with an audit of the data: sensors, sampling, labels, missing values, the structure of the data by specimen, machine or time, and how future data will differ from past data. We ask for a representative sample and a description of the process. Validation is designed around that structure to avoid leakage, and results are compared with simple baselines. Reports give performance with uncertainty, failure cases and the conditions under which the model should not be trusted. No performance figures are promised in advance.
Frequently asked questions
How is this different from the communication-systems signal processing service?
This service targets sensor and measurement data. The communication-systems service concerns modulation, channels and receivers.
Can you analyse vibration data from my machine?
Yes, given the data and the machine description, such as speeds, bearings and gear data. Conclusions depend on data quality and available reference information.
Do you help choose sensors and sampling rates?
We can advise on requirements based on the phenomena of interest, though selection of products is for you.
Can you process biomedical or acoustic signals?
Signal processing methods apply broadly. Domain-specific interpretation should involve a subject expert, and clinical conclusions are outside scope.
Can you build real-time processing?
The service focuses on analysis and algorithm design. Implementation can be considered within an agreed scope.
How is this service scoped and priced?
After your enquiry we review what you need, ask for missing information and send a written scope and quotation. Price depends on complexity, deliverables and timeline. See the project workflow.