AI Defect Detection via Mobile App Camera

Standard machine vision systems (Cognex, Keyence) are effective but expensive and rigidly tied to the line. A mobile inspector based on AI is a flexible alternative: cheaper, more mobile, no equipment rearrangement required. However, new engineering challenges arise: unstable lighting, variable distance to the object, vibration from handheld capture. We solve them with on-device AI tailored to your process. With over 5 years of experience in industrial computer vision and 30+ implementations on production lines, savings of up to 40% compared to off-the-shelf systems are not uncommon. Contact us for a consultation — we'll help you choose the optimal solution.

AI Defect Detection via Mobile App Camera: Technical Challenges

Every millisecond counts, and internet in the workshop is a luxury. On-device inference eliminates latency and network dependency. Model size is limited by device RAM, speed by throughput requirements.

// iOS: industrial defect detection via CoreML
class DefectDetectionEngine {

    private let model: VNCoreMLModel
    private var confidenceThreshold: Float = 0.5   // adjustable on the test stand
    private var iouThreshold: Float = 0.45

    // Dedicated queue for stable FPS
    private let inferenceQueue = DispatchQueue(
        label: "defect.inference",
        qos: .userInteractive
    )

    func analyze(sampleBuffer: CMSampleBuffer) async throws -> [DefectDetection] {
        guard let pixelBuffer = CMSampleBufferGetImageBuffer(sampleBuffer) else {
            throw DefectError.invalidFrame
        }

        return try await withCheckedThrowingContinuation { continuation in
            inferenceQueue.async {
                let request = VNCoreMLRequest(model: self.model) { req, error in
                    if let error = error {
                        continuation.resume(throwing: error)
                        return
                    }
                    let detections = (req.results as? [VNRecognizedObjectObservation])?
                        .filter { $0.confidence >= self.confidenceThreshold }
                        .map { obs in
                            DefectDetection(
                                type: DefectType(rawValue: obs.labels.first?.identifier ?? "") ?? .unknown,
                                confidence: obs.confidence,
                                boundingBox: obs.boundingBox,  // normalized [0,1]
                                severity: self.classifySeverity(obs)
                            )
                        } ?? []
                    continuation.resume(returning: detections)
                }
                request.imageCropAndScaleOption = .scaleFill

                let handler = VNImageRequestHandler(cvPixelBuffer: pixelBuffer)
                try? handler.perform([request])
            }
        }
    }
}

Performance. YOLOv8n in CoreML on iPhone 14: ~15 ms per inference (66 FPS potential). YOLOv8s: ~25 ms. For a line requiring >30 frames/sec with handheld capture — choose the n-variant.

Image Stabilization for Handheld Scanning — AI Defect Detection

Hand vibration is the enemy of small defects. Several techniques: frame buffering with sharpness selection (Laplacian), automatic exposure, optical stabilization if available.

// Frame buffering + selecting the sharpest frame
class StabilizedFrameSelector {
    private var frameBuffer: RingBuffer<CMSampleBuffer> = RingBuffer(capacity: 8)
    private var sharpnessScores: [Float] = []

    func addFrame(_ buffer: CMSampleBuffer) {
        let sharpness = computeLaplacianVariance(buffer)
        frameBuffer.push(buffer)
        sharpnessScores.append(sharpness)
    }

    // For analysis — take the frame with peak sharpness from the last N frames
    var bestFrame: CMSampleBuffer? {
        guard let maxIdx = sharpnessScores.indices.max(by: { sharpnessScores[$0] < sharpnessScores[$1] }) else { return nil }
        return frameBuffer[maxIdx]
    }
}

Also: AVCaptureDevice.activeVideoMinFrameDuration + exposureMode = .continuousAutoExposure + stabilization via videoStabilizationMode = .cinematic.

How to Fine-Tune the Model on Real Data?

Ready-made datasets for specific production are unavailable. You need your own labeling. The process:

1. Capture 200–500 samples on the production line (normal + defective)
2. Label in Label Studio or CVAT (bounding boxes + defect classes)
3. Augmentation: brightness ±30%, rotation ±15°, horizontal flip, Gaussian noise — simulating real shooting conditions
4. Train YOLOv8s/m (depending on speed requirements)
5. Convert to CoreML (.mlpackage) or TFLite
6. Iterative fine-tuning on production errors — every 2–4 weeks

# Fine-tuning on new production data
from ultralytics import YOLO

model = YOLO("defect_detection_v2.pt")  # previous version as base

results = model.train(
    data="production_defects.yaml",
    epochs=50,
    imgsz=640,
    batch=16,
    lr0=0.001,           # lower LR for fine-tuning
    freeze=10,           # freeze first 10 layers of backbone
    augment=True,
    hsv_h=0.015,
    hsv_s=0.7,
    degrees=10.0,
    translate=0.1,
    scale=0.5,
    mosaic=1.0
)

Every 2–4 weeks we fine-tune the model on new production data, increasing accuracy up to 98%. Contact us for an audit of your data — we'll estimate the required labeling volume.

How Do We Integrate the Solution with Your Systems?

The mobile inspector must record results in the MES/ERP system. We use an offline-first approach: first local storage in Room/CoreData, then synchronization via REST/GraphQL when a network connection appears. This guarantees no inspection is lost.

// Android: sending inspection result
data class InspectionResult(
    val productId: String,
    val batchId: String,
    val inspectorId: String,
    val timestamp: Instant,
    val detections: List<DefectDetection>,
    val verdict: InspectionVerdict,   // PASS, FAIL, REVIEW
    val imageUrl: String,             // saved photo with annotations
    val deviceId: String
)

suspend fun submitInspection(result: InspectionResult) {
    // First — local queue (production may be without Wi-Fi)
    localQueue.enqueue(result)
    // Sync when network appears
    syncManager.triggerSync()
}

Offline-first is critically important: workshop Wi-Fi is unstable, loss of an inspection result is unacceptable.

Challenge: Specifics of Industrial Quality Control

Defects in production vary drastically by industry. Below are examples of the most common cases.

For each industry — its own model. A universal defect model doesn't work: what is a defect on a PCB may be normal on metal.

What Does Our Work Include?

All components are configured for your production processes.

Timeline Estimates

MVP with a baseline model (200–300 labeled samples), on-device inference, local inspection history — 3–4 weeks. Full system with a fine-tuned model for a specific production process, image stabilization, offline-first sync with MES/ERP, defect statistics dashboard, and support for iOS + Android — 2–3 months. Exact cost is calculated individually after an audit of your production.

Payback for such a solution is less than 6 months due to reduced scrap and faster inspection. Contact us for a free evaluation of your project. Order a pilot implementation today.

According to Apple's documentation, on-device inference ensures privacy and speed.

Machine Learning in Mobile Apps: CoreML, TFLite, and On-Device Models

We distinguish two fundamentally different approaches: an app with on-device AI and an app that simply calls a cloud API. The former works without internet, does not send user data to third-party servers, and responds within 50 milliseconds. The latter depends on network latency and pricing plans. Choosing the architecture is a key step that directly affects cost, privacy, and user experience in machine learning in mobile apps. Our experience shows that in 70% of projects, on-device inference is cheaper in the long run due to eliminating server costs.

How to Choose Between CoreML and TFLite for On-Device Inference?

CoreML — Apple's native framework for running ML models on device. Supports Neural Engine (starting with A11 Bionic), GPU, and CPU as fallback. Models are converted to .mlmodel format via coremltools from PyTorch, ONNX, or TensorFlow. Conversion is not always trivial: custom layers require implementing MLCustomLayer, and INT8 quantization can sometimes noticeably reduce accuracy on specific data. We ensure the final model passes validation on real data before and after conversion.

TensorFlow Lite — cross-platform alternative for Android and Flutter. On Android it uses NNAPI (Neural Networks API) for hardware acceleration — since Android 10 NNAPI is more stable; before that it's better to explicitly use GPU delegate via GpuDelegate. A typical mistake: the model is trained on normalized data in range [0,1], but the app feeds [0,255] — inference runs but produces meaningless results without any error. We include an automatic input data validation module in the SDK.

For image classification, object detection, and segmentation tasks, ready-to-use optimized models are available. YOLOv8 in CoreML format runs detection on a 640×640 frame in 15–20 ms on iPhone 14 Neural Engine. MobileNetV3 on TFLite with GPU delegate runs around 8 ms on Pixel 7 for classification.

What If You Need Text Generation On-Device?

Running small language models on device has become a reality in the last few years. Apple Intelligence uses its own models via Private Cloud Compute, but for third-party developers other paths are available.

llama.cpp with Metal backend on iOS is a working approach for phi-3-mini (3.8B parameters, 4-bit quantization, ~2.3 GB). Inference: 15–25 tokens/second on iPhone 15 Pro. For integration in Swift, use the Swift Package llama.swift or a wrapper via C interface llama.h. The binary is not bundled with the app — the model is downloaded on first launch and stored in Application Support. Our certified developers configure incremental download to avoid blocking the first launch.

On Android, the analog is Google AI Edge (formerly MediaPipe LLM Inference API) supporting Gemma-2B. It works via GPU delegate, on Tensor G3 chip Pixel 8 Pro — about 20 tokens/second.

Limitations are real: models larger than 4B parameters are still slow on mobile devices. For complex reasoning tasks, on-device LLM falls behind GPT-4o in quality. A hybrid approach — on-device for short tasks and private data, cloud for complex queries — is often optimal. We will evaluate your case and propose a balance of performance and privacy — contact us.

How Does On-Device Inference Compare to Cloud in Terms of Cost and Performance?

On-device inference is typically 10x cheaper per request than cloud APIs for image recognition tasks, while also eliminating latency variability and privacy risks. The table below summarizes the trade-offs.

For an app with 100k MAU running 10 image recognitions per user per month, on-device inference can save up to $5,000 monthly compared to cloud API. Get a free consultation on your ML architecture today.

Integrating OpenAI API and Other Cloud Models

For scenarios where cloud inference is acceptable, integrating OpenAI, Anthropic, or Google Gemini is an HTTP client + streaming SSE. In Swift, AsyncThrowingStream is convenient for streaming responses. In Kotlin, use Flow.

Critically: API keys must never be stored in the app bundle. Even an obfuscated key can be extracted from the IPA in 10 minutes using strings or frida. Correct architecture: mobile app → your own backend → OpenAI API. The backend controls rate limiting, logs requests, and protects the key.

What Is Included in the Work (Deliverables)

• Trained and quantized model for the target device (documentation with metrics)
• SDK for integration (Swift/Kotlin/Flutter) with call examples
• Performance tests on 3–5 real devices
• Instructions for OTA model updates
• Support during App Store / Google Play moderation (compliance with Guidelines 4.2, 5.1)
• 2 weeks of technical support after release

Typical Project Pipeline

1. Task analysis — measure latency, privacy, size, supported devices.
2. Model prototyping — in Python, evaluate accuracy on target data.
3. Conversion and quantization — for CoreML/TFLite with validation.
4. Integration into the app — model wrapped in a service layer (easy to swap CoreML ↔ TFLite ↔ cloud).
5. Testing — on real devices, measure FPS, RAM, battery.
6. Deployment — via TestFlight / Firebase App Distribution, monitor metrics.

Timelines: integration of a ready CoreML/TFLite model — 1–2 weeks, development of a custom model with mobile optimization — from 6 weeks, on-device LLM chat with personalization — 4–8 weeks.

Why We Take on Complex Cases?

10+ years of experience in mobile development, 50+ implemented AI/ML solutions, guarantee of compatibility with current iOS and Android versions. All projects undergo code review and load testing. The cost includes preparation of moderation documentation and training of your team.

Contact us — we will help you choose the architecture and implement ML in your app turnkey. Order an audit of your existing solution — we will assess the potential for server cost savings free of charge. In some projects, savings can reach significant amounts per month.